U.S. patent number 8,026,951 [Application Number 11/670,478] was granted by the patent office on 2011-09-27 for image processing device and method, recording medium, and program.
This patent grant is currently assigned to Sony Corporation. Invention is credited to Naoki Fujiwara, Junichi Ishibashi, Tetsujiro Kondo, Toru Miyake, Takahiro Nagano, Takashi Sawao, Seiji Wada.
United States Patent |
8,026,951 |
Kondo , et al. |
September 27, 2011 |
Image processing device and method, recording medium, and
program
Abstract
An image processing device, method, recording medium, and
program where the device includes a simple-type angle detecting
unit simply detects the angle as continuity using correlation from
an input image. The device includes an input configured to input
image data made up of a plurality of pixels acquired by real world
light signals being cast upon a plurality of detecting elements,
and a real world estimating unit configured to estimate light
signals being cast in an optical low-pass filter.
Inventors: |
Kondo; Tetsujiro (Tokyo,
JP), Nagano; Takahiro (Kanagawa, JP),
Ishibashi; Junichi (Saitama, JP), Sawao; Takashi
(Tokyo, JP), Fujiwara; Naoki (Tokyo, JP),
Wada; Seiji (Kanagawa, JP), Miyake; Toru (Tokyo,
JP) |
Assignee: |
Sony Corporation (Tokyo,
JP)
|
Family
ID: |
32923395 |
Appl.
No.: |
11/670,478 |
Filed: |
February 2, 2007 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20070127838 A1 |
Jun 7, 2007 |
|
Related U.S. Patent Documents
|
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
Issue Date |
|
|
10545081 |
|
7561188 |
|
|
|
PCT/JP2004/001579 |
Feb 13, 2004 |
|
|
|
|
Foreign Application Priority Data
|
|
|
|
|
Feb 28, 2003 [JP] |
|
|
2003-052272 |
|
Current U.S.
Class: |
348/222.1;
382/278; 382/275; 382/300; 382/276 |
Current CPC
Class: |
G06T
7/12 (20170101); G06T 7/74 (20170101); G06T
7/66 (20170101); G06V 10/443 (20220101); G06T
1/00 (20130101); G06K 9/4609 (20130101); G06T
7/77 (20170101) |
Current International
Class: |
H04N
5/228 (20060101); G06K 9/64 (20060101); G06K
9/32 (20060101); G06K 9/40 (20060101); G06K
9/36 (20060101) |
Field of
Search: |
;382/276-308
;348/207.99-376 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
1061473 |
|
Dec 2000 |
|
EP |
|
1164545 |
|
Dec 2001 |
|
EP |
|
2000-201283 |
|
Jul 2000 |
|
JP |
|
2001-84368 |
|
Mar 2001 |
|
JP |
|
2001-250119 |
|
Sep 2001 |
|
JP |
|
2002-152761 |
|
May 2002 |
|
JP |
|
2002-288652 |
|
Oct 2002 |
|
JP |
|
2003-52051 |
|
Feb 2003 |
|
JP |
|
Other References
US. Appl. No. 10/543,839, filed Jul. 29, 2005, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,486, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,734, filed Feb 2, 2007, Kondo et al. cited by
other .
U.S. Appl. No. 11/626,662, filed Jan. 24, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/627,155, filed Jan. 25, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/627,195, filed Jan. 25, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/627,230, filed Jan. 25, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/627,243, filed Jan. 25, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,754, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,732, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,785, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11,670,763, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,776, filed Feb. 2, 2007, Kondo et al. cited
by other .
U.S. Appl. No. 11/670,795, filed Feb. 2, 2007, Kondo et al. cited
by other.
|
Primary Examiner: Chan; Jason
Assistant Examiner: Prabhakher; Pritham
Attorney, Agent or Firm: Oblon, Spivak, McClelland, Maier
& Neustadt, L.L.P.
Parent Case Text
CROSS REFERENCE TO RELATED APPLICATIONS
This application is a continuation of U.S. application Ser. No.
10/545,081, filed on Aug. 9, 2005, and is based upon and claims the
benefit of priority to International Application No.
PCT/JP04/01579, filed on Feb. 13, 2004 and from the prior Japanese
Patent Application No. 2003-052272 filed on Feb. 28, 2003. The
entire contents of each of these documents are incorporated herein
by reference.
Claims
The invention claimed is:
1. An image processing device comprising: an input configured to
input image data made up of a plurality of pixels acquired by real
world light signals being cast upon a plurality of detecting
elements, each having temporal integration effects, via an optical
low-pass filter, of which a part of continuity of said real world
light signals have been lost; and a real world estimating unit
configured to estimate light signals being cast in said optical
low-pass filter, with consideration for light signals being
scattered and integrated in at least a one-dimensional direction
out of spatial directions by said optical low-pass filter, said
real world estimating unit configured to generate functions
approximating said real world light signals, by estimating a
plurality of real world functions, such that pixel values of pixels
of interest, corresponding to a position in two different spatial
directions out of the spatial directions of said image data, are
acquired by integrating in the two different spatial directions
each of the plurality of real world functions corresponding to the
plurality of light signals scattered in a spatial direction by said
optical low-pass filter to generate a plurality of integrated real
world functions, the real world estimating unit configured to
compute pixel values by adding together contributions from at least
two of the plurality of integrated real world functions.
2. The image processing device according to claim 1, further
comprising: an image data continuity detector configured to detect
image data continuity; wherein said real world estimating unit is
configured to generate functions approximating said real world
light signals by estimating said plurality of real world functions,
assuming that the pixel values of pixels of interest, corresponding
to the position in at least the one-dimensional direction out of
the spatial directions of said image data, are the pixel values
acquired by said integration of the plurality of real world
functions corresponding to said optical low-pass filter in at least
the one-dimensional direction, based on the continuity detected by
said image data continuity detector.
3. The image processing device according to claim 2, further
comprising: a pixel value generator configured to generate pixel
values corresponding to pixels of a desired size by integrating
said real world functions which are estimated by said real world
estimating unit in said at least one-dimensional direction in
desired increments.
4. The image processing device according to claim 2, wherein the
real world estimating unit is configured to add together
contributions from at least two of the plurality of real world
functions in proportion to a gradient of the continuity detected by
said image data continuity detector to calculate pixel values.
5. An image processing method comprising: inputting image data made
up of a plurality of pixels acquired by real world light signals
being cast upon a plurality of detecting elements, each having
temporal integration effects, via an optical low-pass filter, of
which a part of continuity of said real world light signals have
been lost; and estimating light signals being cast in said optical
low-pass filter, with consideration for light signals being
scattered and integrated in at least a one-dimensional direction
out of spatial directions by said optical low-pass filter, said
estimating including generating functions approximating said real
world light signals, by estimating a plurality of real world
functions, such that pixel values of pixels of interest,
corresponding to a position in two different spatial directions out
of the spatial directions of said image data, are acquired by
integrating in the two different spatial directions each of the
plurality of real world functions corresponding to the plurality of
light signals scattered in a spatial direction by said optical
low-pass filter to generate a plurality of integrated real world
functions, the estimating including computing pixel values by
adding together contributions from at least two of the plurality of
integrated real world functions.
6. The image processing method according to claim 5, further
comprising: detecting image data continuity, wherein said
estimating includes generating functions approximating said real
world light signals by estimating said plurality of real world
functions, assuming that the pixel values of pixels of interest,
corresponding to the position in at least the one-dimensional
direction out of the spatial directions of said image data, are the
pixel values acquired by said integration of the plurality of real
world functions corresponding to said optical low-pass filter in at
least the one-dimensional direction, based on the continuity
detected by said detecting.
7. The image processing method according to claim 6, wherein the
estimating includes adding together contributions from at least two
of the plurality of real world functions in proportion to a
gradient of the continuity detected by said image data continuity
detector to calculate pixel values.
Description
TECHNICAL FIELD
The present invention relates to an image processing device and
method, a recording medium, and a program, and particularly relates
to an image processing device and method, recording medium, and
program, taking into consideration the real world where data has
been acquired.
BACKGROUND ART
Technology for detecting phenomena in the actual world (real world)
with sensor and processing sampling data output from the sensors is
widely used. For example, image processing technology wherein the
actual world is imaged with an imaging sensor and sampling data
which is the image data is processed, is widely employed.
Also, Japanese Unexamined Patent Application Publication No.
2001-250119 discloses having second dimensions with fewer
dimensions than first dimensions obtained by detecting with sensors
first signals, which are signals of the real world having first
dimensions, obtaining second signals including distortion as to the
first signals, and performing signal processing based on the second
signals, thereby generating third signals with alleviated
distortion as compared to the second signals.
However, signal processing for estimating the first signals from
the second signals had not been thought of to take into
consideration the fact that the second signals for the second
dimensions with fewer dimensions than first dimensions wherein a
part of the continuity of the real world signals is lost, obtained
by first signals which are signals of the real world which has the
first dimensions, have the continuity of the data corresponding to
the stability of the signals of the real world which has been
lost.
DISCLOSURE OF INVENTION
The present invention has been made in light of such a situation,
and it is an object thereof to take into consideration the real
world where data was acquired, and to obtain processing results
which are more accurate and more precise as to phenomena in the
real world.
The image processing device according to the present invention
includes: first angle detecting means for detecting an angle
corresponding to the reference axis of continuity of image data in
image data made up of a plurality of pixels acquired by real world
light signals being cast upon a plurality of detecting elements
each having spatio-temporal integration effects, of which a part of
continuity of said real world light signals have been lost, using
matching processing; second angle detecting means for detecting the
angle using statistical processing based on the image data within a
predetermined region corresponding to the angle detected by the
first angle detecting means; and actual world estimating means for
estimating the light signals by estimating the lost continuity of
the real world light signals based on the angle detected by the
second angle detecting means.
The first angle detecting means may include: pixel detecting means
for detecting an image block centered on a plurality of pixels
adjacent to the straight line of each angle on the basis of a pixel
of interest within the image data; and correlation detecting means
for detecting correlation of image blocks detected by the pixel
detecting means; wherein the angle as to the reference axis of
continuity of the image data is detected according to the value of
correlation of the image blocks detected by the correlation
detecting means.
The second angle detecting means may further include: a plurality
of statistical processing means; wherein the angle may be detected
using one statistical processing means of the plurality of
statistical processing means according to the angle detected by the
first angle detecting means.
One statistical processing means of the plurality of statistical
processing means may further include: dynamic range detecting means
for detecting a dynamic range, which is difference between the
maximum value and the minimum value of the pixel values of the
pixels within the predetermined region; difference value detecting
means for detecting the difference value between adjacent pixels in
the direction according to activity within the predetermined
region; and statistical angle detecting means for statistically
detecting an angle as to the reference axis of continuity of image
data corresponding to the lost continuity of the real world light
signals, according to the dynamic range and the difference
value.
One statistical processing means of the plurality of statistical
processing means may include: score detecting means for taking the
number of pixels of which the correlation value as to the pixel
value of another pixel within the predetermined region is equal to
or greater than a threshold value as a score corresponding to the
pixel of interest; and statistical angle detecting means for
statistically detecting an angle as to the reference axis of
continuity of the image data by detecting a regression line based
on the score of each pixel of interest detected by the score
detecting means.
The image processing method according to the present invention
includes: a first angle detecting step for detecting an angle
corresponding to the reference axis of continuity of image data in
image data made up of a plurality of pixels acquired by real world
light signals being cast upon a plurality of detecting elements
each having spatio-temporal integration effects, of which a part of
continuity of the real world light signals have been lost, using
matching processing; a second angle detecting step for detecting
the angle using statistical processing based on the image data
within a predetermined region corresponding to the angle detected
in the first angle detecting step; and an actual world estimating
step for estimating the light signals by estimating the lost
continuity of the real world light signals based on the angle
detected in the second angle detecting step.
The program of the recording medium according to the present
invention is a program that can be read by a computer which
executes processing including: a first angle detecting step for
detecting an angle corresponding to the reference axis of
continuity of image data in image data made up of a plurality of
pixels acquired by real world light signals being cast upon a
plurality of detecting elements each having spatio-temporal
integration effects, of which a part of continuity of the real
world light signals have been lost, using matching processing; a
second angle detecting step for detecting the angle using
statistical processing based on the image data within a
predetermined region corresponding to the angle detected in the
first angle detecting step; and an actual world estimating step for
estimating the light signals by estimating the lost continuity of
the real world light signals based on the angle detected in the
second angle detecting step.
The program according to the present invention causes a computer to
execute processing including: a first angle detecting step for
detecting an angle corresponding to the reference axis of
continuity of image data in image data made up of a plurality of
pixels acquired by real world light signals being cast upon a
plurality of detecting elements each having spatio-temporal
integration effects, of which a part of continuity of the real
world light signals have been lost, using matching processing; a
second angle detecting step for detecting the angle using
statistical processing based on the image data within a
predetermined region corresponding to the angle detected in the
first angle detecting step; and an actual world estimating step for
estimating the light signals by estimating the lost continuity of
the real world light signals based on the angle detected in the
second angle detecting step.
With the image processing device and method, and program, according
to the present invention, an angle corresponding to the reference
axis of continuity of image data in image data made up of a
plurality of pixels acquired by real world light signals being cast
upon a plurality of detecting elements each having spatio-temporal
integration effects, of which a part of continuity of said real
world light signals have been lost, is detected with matching
processing, an angle is detected with statistical processing based
on the image data within a predetermined region corresponding to
the detected angle, the light signals are estimated by estimating
the lost continuity of the real world light signals based on the
angle detected with the statistical processing.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a diagram illustrating the principle of the present
invention.
FIG. 2 is a block diagram illustrating an example of a
configuration of a signal processing device 4.
FIG. 3 is a block diagram illustrating a signal processing device
4.
FIG. 4 is a diagram illustrating the principle of processing of a
conventional image processing device 121.
FIG. 5 is a diagram for describing the principle of processing of
the image processing device 4.
FIG. 6 is a diagram for describing the principle of the present
invention in greater detail.
FIG. 7 is a diagram for describing the principle of the present
invention in greater detail.
FIG. 8 is a diagram describing an example of the placement of
pixels on an image sensor.
FIG. 9 is a diagram for describing the operations of a detecting
device which is a CCD.
FIG. 10 is a diagram for describing the relationship between light
cast into detecting elements corresponding to pixel D through pixel
F, and pixel values.
FIG. 11 is a diagram for describing the relationship between the
passage of time, light cast into a detecting element corresponding
to one pixel, and pixel values.
FIG. 12 is a diagram illustrating an example of an image of a
linear-shaped object in the actual world 1.
FIG. 13 is a diagram illustrating an example of pixel values of
image data obtained by actual image-taking.
FIG. 14 is a schematic diagram of image data.
FIG. 15 is a diagram illustrating an example of an image of an
actual world 1 having a linear shape of a single color which is a
different color from the background.
FIG. 16 is a diagram illustrating an example of pixel values of
image data obtained by actual image-taking.
FIG. 17 is a schematic diagram of image data.
FIG. 18 is a diagram for describing the principle of the present
invention.
FIG. 19 is a diagram for describing the principle of the present
invention.
FIG. 20 is a diagram for describing an example of generating
high-resolution data 181.
FIG. 21 is a diagram for describing approximation by a model
161.
FIG. 22 is a diagram for describing estimation of the model 161
with M pieces of data 162.
FIG. 23 is a diagram for describing the relationship between
signals of the actual world 1 and data 3.
FIG. 24 is a diagram illustrating an example of data 3 of interest
at the time of creating an Expression.
FIG. 25 is a diagram for describing signals for two objects in the
actual world, and values belonging to a mixed region at the time of
creating an expression.
FIG. 26 is a diagram for describing continuity represented by
Expression (18), Expression (19), and Expression (22).
FIG. 27 is a diagram illustrating an example of M pieces of data
extracted from data.
FIG. 28 is a diagram for describing a region where a pixel value,
which is data 3, is obtained.
FIG. 29 is a diagram for describing approximation of the position
of a pixel in the space-time direction.
FIG. 30 is a diagram for describing integration of signals of the
actual world 1 in the time direction and two-dimensional spatial
direction, in the data 3.
FIG. 31 is a diagram for describing an integration region at the
time of generating high-resolution data 181 with higher resolution
in the spatial direction.
FIG. 32 is a diagram for describing an integration region at the
time of generating high-resolution data 181 with higher resolution
in the time direction.
FIG. 33 is a diagram for describing an integration region at the
time of generating high-resolution data 181 with blurring due to
movement having been removed.
FIG. 34 is a diagram for describing an integration region at the
time of generating high-resolution data 181 with higher resolution
in the time-space direction.
FIG. 35 is a diagram illustrating the original image of the input
image.
FIG. 36 is a diagram illustrating an example of an input image.
FIG. 37 is a diagram illustrating an image obtained by applying
conventional class classification adaptation processing.
FIG. 38 is a diagram illustrating results of detecting a region
with a fine line.
FIG. 39 is a diagram illustrating an example of an output image
output from a signal processing device 4.
FIG. 40 is a flowchart for describing signal processing with the
signal processing device 4.
FIG. 41 is a block diagram illustrating the configuration of a data
continuity detecting unit.
FIG. 42 is a diagram illustrating an image in the actual world 1
with a fine line in front of the background.
FIG. 43 is a diagram for describing approximation of a background
with a plane.
FIG. 44 is a diagram illustrating the cross-sectional shape of
image data regarding which the image of a fine line has been
projected.
FIG. 45 is a diagram illustrating the cross-sectional shape of
image data regarding which the image of a fine line has been
projected.
FIG. 46 is a diagram illustrating the cross-sectional shape of
image data regarding which the image of a fine line has been
projected.
FIG. 47 is a diagram for describing the processing for detecting a
peak and detecting of monotonous increase/decrease regions.
FIG. 48 is a diagram for describing the processing for detecting a
fine line region wherein the pixel value of the peak exceeds a
threshold, while the pixel value of the adjacent pixel is equal to
or below the threshold value.
FIG. 49 is a diagram representing the pixel value of pixels arrayed
in the direction indicated by dotted line AA' in FIG. 48.
FIG. 50 is a diagram for describing processing for detecting
continuity in a monotonous increase/decrease region.
FIG. 51 is a diagram illustrating an example of an image regarding
which a continuity component has been extracted by approximation on
a plane.
FIG. 52 is a diagram illustrating results of detecting regions with
monotonous decrease.
FIG. 53 is a diagram illustrating regions where continuity has been
detected.
FIG. 54 is a diagram illustrating pixel values at regions where
continuity has been detected.
FIG. 55 is a diagram illustrating an example of other processing
for detecting regions where an image of a fine line has been
projected.
FIG. 56 is a flowchart for describing continuity detection
processing.
FIG. 57 is a diagram for describing processing for detecting
continuity of data in the time direction.
FIG. 58 is a block diagram illustrating the configuration of a
non-continuity component extracting unit 201.
FIG. 59 is a diagram for describing the number of time of
rejections.
FIG. 60 is a diagram illustrating an example of an input image.
FIG. 61 is a diagram illustrating an image wherein standard error
obtained as the result of planar approximation without rejection is
taken as pixel values.
FIG. 62 is a diagram illustrating an image wherein standard error
obtained as the result of planar approximation with rejection is
taken as pixel values.
FIG. 63 is a diagram illustrating an image wherein the number of
times of rejection is taken as pixel values.
FIG. 64 is a diagram illustrating an image wherein the gradient of
the spatial direction X of a plane is taken as pixel values.
FIG. 65 is a diagram illustrating an image wherein the gradient of
the spatial direction Y of a plane is taken as pixel values.
FIG. 66 is a diagram illustrating an image formed of planar
approximation values.
FIG. 67 is a diagram illustrating an image formed of the difference
between planar approximation values and pixel values.
FIG. 68 is a flowchart describing the processing for extracting the
non-continuity component.
FIG. 69 is a flowchart describing the processing for extracting the
continuity component.
FIG. 70 is a flowchart describing other processing for extracting
the continuity component.
FIG. 71 is a flowchart describing still other processing for
extracting the continuity component.
FIG. 72 is a block diagram illustrating another configuration of a
data continuity detecting unit 101.
FIG. 73 is a diagram for describing the activity on an input image
having data continuity.
FIG. 74 is a diagram for describing a block for detecting
activity.
FIG. 75 is a diagram for describing the angle of data continuity as
to activity.
FIG. 76 is a block diagram illustrating a detailed configuration of
the data continuity detecting unit 101.
FIG. 77 is a diagram describing a set of pixels.
FIG. 78 is a diagram describing the relation between the position
of a pixel set and the angle of data continuity.
FIG. 79 is a flowchart for describing processing for detecting data
continuity.
FIG. 80 is a diagram illustrating a set of pixels extracted when
detecting the angle of data continuity in the time direction and
space direction.
FIG. 81 is a block diagram illustrating another further detailed
configuration of the data continuity detecting unit 101.
FIG. 82 is a diagram for describing a set of pixels made up of
pixels of a number corresponding to the range of angle of set
straight lines.
FIG. 83 is a diagram describing the range of angle of the set
straight lines.
FIG. 84 is a diagram describing the range of angle of the set
straight lines, the number of pixel sets, and the number of pixels
per pixel set.
FIG. 85 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 86 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 87 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 88 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 89 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 90 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 91 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 92 is a diagram for describing the number of pixel sets and
the number of pixels per pixel set.
FIG. 93 is a flowchart for describing processing for detecting data
continuity.
FIG. 94 is a block diagram illustrating still another configuration
of the data continuity detecting unit 101.
FIG. 95 is a block diagram illustrating a further detailed
configuration of the data continuity detecting unit 101.
FIG. 96 is a diagram illustrating an example of a block.
FIG. 97 is a diagram describing the processing for calculating the
absolute value of difference of pixel values between a block of
interest and a reference block.
FIG. 98 is a diagram describing the distance in the spatial
direction X between the position of a pixel in the proximity of the
pixel of interest, and a straight line having an angle .theta..
FIG. 99 is a diagram illustrating the relationship between the
shift amount .gamma. and angle .theta..
FIG. 100 is a diagram illustrating the distance in the spatial
direction X between the position of a pixel in the proximity of the
pixel of interest and a straight line which passes through the
pixel of interest and has an angle of .theta., as to the shift
amount .gamma..
FIG. 101 is a diagram illustrating reference block wherein the
distance as to a straight line which passes through the pixel of
interest and has an angle of .theta. as to the axis of the spatial
direction X, is minimal.
FIG. 102 is a diagram for describing processing for halving the
range of angle of continuity of data to be detected.
FIG. 103 is a flowchart for describing the processing for detection
of data continuity.
FIG. 104 is a diagram illustrating a block which is extracted at
the time of detecting the angle of data continuity in the space
direction and time direction.
FIG. 105 is a block diagram illustrating the configuration of the
data continuity detecting unit 101 which executes processing for
detection of data continuity, based on components signals of an
input image.
FIG. 106 is a block diagram illustrating the configuration of the
data continuity detecting unit 101 which executes processing for
detection of data continuity, based on components signals of an
input image.
FIG. 107 is a block diagram illustrating still another
configuration of the data continuity detecting unit 101.
FIG. 108 is a diagram for describing the angle of data continuity
with a reference axis as a reference, in the input image.
FIG. 109 is a diagram for describing the angle of data continuity
with a reference axis as a reference, in the input image.
FIG. 110 is a diagram for describing the angle of data continuity
with a reference axis as a reference, in the input image.
FIG. 111 is a diagram illustrating the relationship between the
change in pixel values as to the position of pixels in the spatial
direction, and a regression line, in the input image.
FIG. 112 is a diagram for describing the angle between the
regression line A, and an axis indicating the spatial direction X,
which is a reference axis, for example.
FIG. 113 is a diagram illustrating an example of a region.
FIG. 114 is a flowchart for describing the processing for detection
of data continuity with the data continuity detecting unit 101 of
which the configuration is illustrated in FIG. 107.
FIG. 115 is a block diagram illustrating still another
configuration of the data continuity detecting unit 101.
FIG. 116 is a diagram illustrating the relationship between the
change in pixel values as to the position of pixels in the spatial
direction, and a regression line, in the input image.
FIG. 117 is a diagram for describing the relationship between
standard deviation and a region having data continuity.
FIG. 118 is a diagram illustrating an example of a region.
FIG. 119 is a flowchart for describing the processing for detection
of data continuity with the data continuity detecting unit 101 of
which the configuration is illustrated in FIG. 115.
FIG. 120 is a flowchart for describing other processing for
detection of data continuity with the data continuity detecting
unit 101 of which the configuration is illustrated in FIG. 115.
FIG. 121 is a block diagram illustrating the configuration of the
data continuity detecting unit for detecting the angle of a fine
line or a two-valued edge, as data continuity information, to which
the present invention has been applied.
FIG. 122 is a diagram for describing a detection method for data
continuity information.
FIG. 123 is a diagram for describing a detection method for data
continuity information.
FIG. 124 is a diagram illustrating a further detailed configuration
of the data continuity detecting unit.
FIG. 125 is a diagram for describing horizontal/vertical
determination processing.
FIG. 126 is a diagram for describing horizontal/vertical
determination processing.
FIG. 127A is a diagram for describing the relationship between a
fine line in the real world and a fine line imaged by a sensor.
FIG. 127B is a diagram for describing the relationship between a
fine line in the real world and a fine line imaged by a sensor.
FIG. 127C is a diagram for describing the relationship between a
fine line in the real world and a fine line imaged by a sensor.
FIG. 128A is a diagram for describing the relationship between a
fine line in the real world and the background.
FIG. 128B is a diagram for describing the relationship between a
fine line in the real world and the background.
FIG. 129A is a diagram for describing the relationship between a
fine line in an image imaged by a sensor and the background.
FIG. 129B is a diagram for describing the relationship between a
fine line in an image imaged by a sensor and the background.
FIG. 130A is a diagram for describing an example of the
relationship between a fine line in an image imaged by a sensor and
the background.
FIG. 130B is a diagram for describing an example of the
relationship between a fine line in an image imaged by a sensor and
the background.
FIG. 131A is a diagram for describing the relationship between a
fine line in an image in the real world and the background.
FIG. 131B is a diagram for describing the relationship between a
fine line in an image in the real world and the background.
FIG. 132A is a diagram for describing the relationship between a
fine line in an image imaged by a sensor and the background.
FIG. 132B is a diagram for describing the relationship between a
fine line in an image imaged by a sensor and the background.
FIG. 133A is a diagram for describing an example of the
relationship between a fine line in an image imaged by a sensor and
the background.
FIG. 133B is a diagram for describing an example of the
relationship between a fine line in an image imaged by a sensor and
the background.
FIG. 134 is a diagram illustrating a model for obtaining the angle
of a fine line.
FIG. 135 is a diagram illustrating a model for obtaining the angle
of a fine line.
FIG. 136A is a diagram for describing the maximum value and minimum
value of pixel values in a dynamic range block corresponding to a
pixel of interest.
FIG. 136B is a diagram for describing the maximum value and minimum
value of pixel values in a dynamic range block corresponding to a
pixel of interest.
FIG. 137A is a diagram for describing how to obtain the angle of a
fine line.
FIG. 137B is a diagram for describing how to obtain the angle of a
fine line.
FIG. 137C is a diagram for describing how to obtain the angle of a
fine line.
FIG. 138 is a diagram for describing how to obtain the angle of a
fine line.
FIG. 139 is a diagram for describing an extracted block and dynamic
range block.
FIG. 140 is a diagram for describing a least-square solution.
FIG. 141 is a diagram for describing a least-square solution.
FIG. 142A is a diagram for describing a two-valued edge.
FIG. 142B is a diagram for describing a two-valued edge.
FIG. 142C is a diagram for describing a two-valued edge.
FIG. 143A is a diagram for describing a two-valued edge of an image
imaged by a sensor.
FIG. 143B is a diagram for describing a two-valued edge of an image
imaged by a sensor.
FIG. 144A is a diagram for describing an example of a two-valued
edge of an image imaged by a sensor.
FIG. 144B is a diagram for describing an example of a two-valued
edge of an image imaged by a sensor.
FIG. 145A is a diagram for describing a two-valued edge of an image
imaged by a sensor.
FIG. 145B is a diagram for describing a two-valued edge of an image
imaged by a sensor.
FIG. 146 is a diagram illustrating a model for obtaining the angle
of a two-valued edge.
FIG. 147A is a diagram illustrating a method for obtaining the
angle of a two-valued edge.
FIG. 147B is a diagram illustrating a method for obtaining the
angle of a two-valued edge.
FIG. 147C is a diagram illustrating a method for obtaining the
angle of a two-valued edge.
FIG. 148 is a diagram illustrating a method for obtaining the angle
of a two-valued edge.
FIG. 149 is a flowchart for describing the processing for detecting
the angle of a fine line or a two-valued edge along with data
continuity.
FIG. 150 is a flowchart for describing data extracting
processing.
FIG. 151 is a flowchart for describing addition processing to a
normal equation.
FIG. 152A is a diagram for comparing the gradient of a fine line
obtained by application of the present invention, and the angle of
a fine line obtained using correlation.
FIG. 152B is a diagram for comparing the gradient of a fine line
obtained by application of the present invention, and the angle of
a fine line obtained using correlation.
FIG. 153A is a diagram for comparing the gradient of a two-valued
edge obtained by application of the present invention, and the
angle of a fine line obtained using correlation.
FIG. 153B is a diagram for comparing the gradient of a two-valued
edge obtained by application of the present invention, and the
angle of a fine line obtained using correlation.
FIG. 154 is a block diagram illustrating the configuration of the
data continuity detecting unit for detecting a mixture ratio under
application of the present invention as data continuity
information.
FIG. 155A is a diagram for describing how to obtain the mixture
ratio.
FIG. 155B is a diagram for describing how to obtain the mixture
ratio.
FIG. 155C is a diagram for describing how to obtain the mixture
ratio.
FIG. 156 is a flowchart for describing processing for detecting the
mixture ratio along with data continuity.
FIG. 157 is a flowchart for describing addition processing to a
normal equation.
FIG. 158A is a diagram illustrating an example of distribution of
the mixture ratio of a fine line.
FIG. 158B is a diagram illustrating an example of distribution of
the mixture ratio of a fine line.
FIG. 159A is a diagram illustrating an example of distribution of
the mixture ratio of a two-valued edge.
FIG. 159B is a diagram illustrating an example of distribution of
the mixture ratio of a two-valued edge.
FIG. 160 is a diagram for describing linear approximation of the
mixture ratio.
FIG. 161A is a diagram for describing a method for obtaining
movement of an object as data continuity information.
FIG. 161B is a diagram for describing a method for obtaining
movement of an object as data continuity information.
FIG. 162A is a diagram for describing a method for obtaining
movement of an object as data continuity information.
FIG. 162B is a diagram for describing a method for obtaining
movement of an object as data continuity information.
FIG. 163A is a diagram for describing a method for obtaining a
mixture ratio according to movement of an object as data continuity
information.
FIG. 163B is a diagram for describing a method for obtaining a
mixture ratio according to movement of an object as data continuity
information.
FIG. 163C is a diagram for describing a method for obtaining a
mixture ratio according to movement of an object as data continuity
information.
FIG. 164 is a diagram for describing linear approximation of the
mixture ratio at the time of obtaining the mixture ratio according
to movement of the object as data continuity information.
FIG. 165 is a block diagram illustrating the configuration of the
data continuity detecting unit for detecting the processing region
under application of the present invention, as data continuity
information.
FIG. 166 is a flowchart for describing the processing for detection
of continuity with the data continuity detecting unit shown in FIG.
165.
FIG. 167 is a diagram for describing the integration range of
processing for detection of continuity with the data continuity
detecting unit shown in FIG. 165.
FIG. 168 is a diagram for describing the integration range of
processing for detection of continuity with the data continuity
detecting unit shown in FIG. 165.
FIG. 169 is a block diagram illustrating another configuration of
the data continuity detecting unit for detecting a processing
region to which the present invention has been applied as data
continuity information.
FIG. 170 is a flowchart for describing the processing for detecting
continuity with the data continuity detecting unit shown in FIG.
169.
FIG. 171 is a diagram for describing the integration range of
processing for detecting continuity with the data continuity
detecting unit shown in FIG. 169.
FIG. 172 is a diagram for describing the integration range of
processing for detecting continuity with the data continuity
detecting unit shown in FIG. 169.
FIG. 173 is a block diagram illustrating the configuration of
another embodiment of the data continuity detecting unit.
FIG. 174 is a block diagram illustrating an example of a
configuration of a simple-type angle detecting unit of the data
continuity detecting unit shown in FIG. 173.
FIG. 175 is a block diagram illustrating an example of a
configuration of a regression-type angle detecting unit of the data
continuity detecting unit shown in FIG. 173.
FIG. 176 is a block diagram illustrating an example of a
configuration of a gradient-type angle detecting unit of the data
continuity detecting unit shown in FIG. 173.
FIG. 177 is a flowchart for describing the processing for detecting
continuity of data with the data continuity detecting unit shown in
FIG. 173.
FIG. 178 is a diagram for describing a method for detecting an
angle corresponding to the angle detected by the simple-type angle
detecting unit.
FIG. 179 is a flowchart for describing the regression-type angle
detecting processing, which is the processing in step S904 in the
flowchart shown in FIG. 177.
FIG. 180 is a diagram for describing pixels serving as a scope
range where the score conversion processing is performed.
FIG. 181 is a diagram for describing pixels serving as a scope
range where the score conversion processing is performed.
FIG. 182 is a diagram for describing pixels serving as a scope
range where the score conversion processing is performed.
FIG. 183 is a diagram for describing pixels serving as a scope
range where the score conversion processing is performed.
FIG. 184 is a diagram for describing pixels serving as a scope
range where the score conversion processing is performed.
FIG. 185 is a block diagram illustrating the configuration of
another embodiment of the data continuity detecting unit.
FIG. 186 is a flowchart for describing the processing for detecting
continuity of data with the data continuity detecting unit shown in
FIG. 185.
FIG. 187 is a block diagram illustrating the configuration of an
actual world estimating unit 102.
FIG. 188 is a diagram for describing the processing for detecting
the width of a fine line in actual world 1 signals.
FIG. 189 is a diagram for describing the processing for detecting
the width of a fine line in actual world 1 signals.
FIG. 190 is a diagram for describing the processing for estimating
the level of a fine line signal in actual world 1 signals.
FIG. 191 is a flowchart for describing processing for estimating
the actual world.
FIG. 192 is a block diagram illustrating another configuration of
the actual world estimating unit 102.
FIG. 193 is a block diagram illustrating the configuration of a
boundary detecting unit 2121.
FIG. 194 is a diagram for describing the processing for calculating
allocation ratio.
FIG. 195 is a diagram for describing the processing for calculating
allocation ratio.
FIG. 196 is a diagram for describing the processing for calculating
allocation ratio.
FIG. 197 is a diagram for describing the process for calculating a
regression line indicating the boundary of monotonous
increase/decrease regions.
FIG. 198 is a diagram for describing the process for calculating a
regression line indicating the boundary of monotonous
increase/decrease regions.
FIG. 199 is a flowchart for describing processing for estimating
the actual world.
FIG. 200 is a flowchart for describing the processing for boundary
detection.
FIG. 201 is a block diagram illustrating the configuration of the
actual world estimating unit which estimates the derivative value
in the spatial direction as actual world estimating
information.
FIG. 202 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 201.
FIG. 203 is a diagram for describing a reference pixel.
FIG. 204 is a diagram for describing the position for obtaining the
derivative value in the spatial direction.
FIG. 205 is a diagram for describing the relationship between the
derivative value in the spatial direction and the amount of
shift.
FIG. 206 is a block diagram illustrating the configuration of the
actual world estimating unit which estimates the gradient in the
spatial direction as actual world estimating information.
FIG. 207 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 206.
FIG. 208 is a diagram for describing processing for obtaining the
gradient in the spatial direction.
FIG. 209 is a diagram for describing processing for obtaining the
gradient in the spatial direction.
FIG. 210 is a block diagram illustrating the configuration of the
actual world estimating unit for estimating the derivative value in
the frame direction as actual world estimating information.
FIG. 211 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 210.
FIG. 212 is a diagram for describing a reference pixel.
FIG. 213 is a diagram for describing the position for obtaining the
derivative value in the frame direction.
FIG. 214 is a diagram for describing the relationship between the
derivative value in the frame direction and the amount of
shift.
FIG. 215 is a block diagram illustrating the configuration of the
actual world estimating unit which estimates the gradient in the
frame direction as actual world estimating information.
FIG. 216 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 215.
FIG. 217 is a diagram for describing processing for obtaining the
gradient in the frame direction.
FIG. 218 is a diagram for describing processing for obtaining the
gradient in the frame direction.
FIG. 219 is a diagram for describing the principle of function
approximation, which is an example of an embodiment of the actual
world estimating unit shown in FIG. 3.
FIG. 220 is a diagram for describing integration effects in the
event that the sensor is a CCD.
FIG. 221 is a diagram for describing a specific example of the
integration effects of the sensor shown in FIG. 220.
FIG. 222 is a diagram for describing a specific example of the
integration effects of the sensor shown in FIG. 220.
FIG. 223 is a diagram representing a fine-line-inclusive actual
world region shown in FIG. 221.
FIG. 224 is a diagram for describing the principle of an example of
an embodiment of the actual world estimating unit shown in FIG. 3,
in comparison with the example shown in FIG. 219.
FIG. 225 is a diagram representing the fine-line-inclusive data
region shown in FIG. 221.
FIG. 226 is a diagram wherein each of the pixel values contained in
the fine-line-inclusive data region shown in FIG. 225 are plotted
on a graph.
FIG. 227 is a diagram wherein an approximation function,
approximating the pixel values contained in the fine-line-inclusive
data region shown in FIG. 226, is plotted on a graph.
FIG. 228 is a diagram for describing the continuity in the spatial
direction which the fine-line-inclusive actual world region shown
in FIG. 221 has.
FIG. 229 is a diagram wherein each of the pixel values contained in
the fine-line-inclusive data region shown in FIG. 225 are plotted
on a graph.
FIG. 230 is a diagram for describing a state wherein each of the
input pixel values indicated in FIG. 229 are shifted by a
predetermined shift amount.
FIG. 231 is a diagram wherein an approximation function,
approximating the pixel values contained in the fine-line-inclusive
data region shown in FIG. 226, is plotted on a graph, taking into
consideration the spatial-direction continuity.
FIG. 232 is a diagram for describing space-mixed region.
FIG. 233 is a diagram for describing an approximation function
approximating actual-world signals in a space-mixed region.
FIG. 234 is a diagram wherein an approximation function,
approximating the actual world signals corresponding to the
fine-line-inclusive data region shown in FIG. 226, is plotted on a
graph, taking into consideration both the sensor integration
properties and the spatial-direction continuity.
FIG. 235 is a block diagram for describing a configuration example
of the actual world estimating unit using, of function
approximation techniques having the principle shown in FIG. 219,
primary polynomial approximation.
FIG. 236 is a flowchart for describing actual world estimation
processing which the actual world estimating unit of the
configuration shown in FIG. 235 executes.
FIG. 237 is a diagram for describing a tap range.
FIG. 238 is a diagram for describing actual world signals having
continuity in the spatial direction.
FIG. 239 is a diagram for describing integration effects in the
event that the sensor is a CCD.
FIG. 240 is a diagram for describing distance in the
cross-sectional direction.
FIG. 241 is a block diagram for describing a configuration example
of the actual world estimating unit using, of function
approximation techniques having the principle shown in FIG. 219,
quadratic polynomial approximation.
FIG. 242 is a flowchart for describing actual world estimation
processing which the actual world estimating unit of the
configuration shown in FIG. 241 executes.
FIG. 243 is a diagram for describing a tap range.
FIG. 244 is a diagram for describing direction of continuity in the
time-spatial direction.
FIG. 245 is a diagram for describing integration effects in the
event that the sensor is a CCD.
FIG. 246 is a diagram for describing actual world signals having
continuity in the spatial direction.
FIG. 247 is a diagram for describing actual world signals having
continuity in the space-time directions.
FIG. 248 is a block diagram for describing a configuration example
of the actual world estimating unit using, of function
approximation techniques having the principle shown in FIG. 219,
cubic polynomial approximation.
FIG. 249 is a flowchart for describing actual world estimation
processing which the actual world estimating unit of the
configuration shown in FIG. 248 executes.
FIG. 250 is a diagram illustrating an example of an input image to
be input in the actual world estimating unit shown in FIG. 3.
FIG. 251 is a diagram illustrating the difference between the
actual world light signal level in the center of the pixel of
interest shown in FIG. 250 and the actual world light signal level
in the cross-sectional direction distance x'.
FIG. 252 is a diagram for describing the cross-sectional direction
distance x'.
FIG. 253 is a diagram for describing the cross-sectional direction
distance x'.
FIG. 254 is a diagram illustrating the cross-sectional direction
distance of each pixel within a block.
FIG. 255 is a diagram illustrating the result of processing without
taking into consideration weight in a normal equation.
FIG. 256 is a diagram illustrating the result of processing with
taking into consideration weight in a normal equation.
FIG. 257 is a diagram illustrating the result of processing without
taking into consideration weight in a normal equation.
FIG. 258 is a diagram illustrating the result of processing with
taking into consideration weight in a normal equation.
FIG. 259 is a diagram for describing the principle of
re-integration, which is an example of an embodiment of the image
generating unit shown in FIG. 3.
FIG. 260 is a diagram for describing an example of an input pixel
and an approximation function for approximation of an actual world
signal corresponding to the input pixel.
FIG. 261 is a diagram for describing an example of creating four
high-resolution pixels in the one input pixel shown in FIG. 260,
from the approximation function shown in FIG. 260.
FIG. 262 is a block diagram for describing a configuration example
of an image generating unit using, of re-integration techniques
having the principle shown in FIG. 259, one-dimensional
re-integration technique.
FIG. 263 is a flowchart for describing the image generating
processing which the image generating unit of the configuration
shown in FIG. 262 executes.
FIG. 264 is a diagram illustrating an example of the original image
of the input image.
FIG. 265 is a diagram illustrating an example of image data
corresponding to the image shown in FIG. 264.
FIG. 266 is a diagram illustrating an example of an input
image.
FIG. 267 is a diagram representing an example of image data
corresponding to the image shown in FIG. 266.
FIG. 268 is a diagram illustrating an example of an image obtained
by subjecting an input image to conventional class classification
adaptation processing.
FIG. 269 is a diagram representing an example of image data
corresponding to the image shown in FIG. 268.
FIG. 270 is a diagram illustrating an example of an image obtained
by subjecting an input image to the one-dimensional re-integration
technique according to the present invention.
FIG. 271 is a diagram illustrating an example of image data
corresponding to the image shown in FIG. 270.
FIG. 272 is a diagram for describing actual-world signals having
continuity in the spatial direction.
FIG. 273 is a block diagram for describing a configuration example
of an image generating unit which uses, of the re-integration
techniques having the principle shown in FIG. 259, a
two-dimensional re-integration technique.
FIG. 274 is a diagram for describing distance in the
cross-sectional direction.
FIG. 275 is a flowchart for describing the image generating
processing which the image generating unit of the configuration
shown in FIG. 273 executes.
FIG. 276 is a diagram for describing an example of an input
pixel.
FIG. 277 is a diagram for describing an example of creating four
high-resolution pixels in the one input pixel shown in FIG. 276,
with the two-dimensional re-integration technique.
FIG. 278 is a diagram for describing the direction of continuity in
the space-time directions.
FIG. 279 is a block diagram for describing a configuration example
of the image generating unit which uses, of the re-integration
techniques having the principle shown in FIG. 259, a
three-dimensional re-integration technique.
FIG. 280 is a flowchart for describing the image generating
processing which the image generating unit of the configuration
shown in FIG. 279 executes.
FIG. 281 is a block diagram illustrating another configuration of
the image generating unit to which the present invention is
applied.
FIG. 282 is a flowchart for describing the processing for image
generating with the image generating unit shown in FIG. 281.
FIG. 283 is a diagram for describing processing of creating a
quadruple density pixel from an input pixel.
FIG. 284 is a diagram for describing the relationship between an
approximation function indicating the pixel value and the amount of
shift.
FIG. 285 is a block diagram illustrating another configuration of
the image generating unit to which the present invention has been
applied.
FIG. 286 is a flowchart for describing the image generating
processing with the image generating unit shown in FIG. 285.
FIG. 287 is a diagram for describing processing of creating a
quadruple density pixel from an input pixel.
FIG. 288 is a diagram for describing the relationship between an
approximation function indicating the pixel value and the amount of
shift.
FIG. 289 is a block diagram for describing a configuration example
of the image generating unit which uses the one-dimensional
re-integration technique in the class classification adaptation
process correction technique, which is an example of an embodiment
of the image generating unit shown in FIG. 3.
FIG. 290 is a block diagram describing a configuration example of
the class classification adaptation processing unit of the image
generating unit shown in FIG. 289.
FIG. 291 is a block diagram illustrating a configuration example of
the class classification adaptation processing unit shown in FIG.
289, and a learning device for determining a coefficient for the
class classification adaptation processing correction unit to use
by way of learning.
FIG. 292 is a block diagram for describing a detailed configuration
example of the learning unit for the class classification
adaptation processing, shown in FIG. 291.
FIG. 293 is a diagram illustrating an example of processing results
of the class classification adaptation processing unit shown in
FIG. 290.
FIG. 294 is a diagram illustrating a difference image between the
prediction image shown in FIG. 293 and an HD image.
FIG. 295 is a diagram plotting each of specific pixel values of the
HD image in FIG. 293, specific pixel values of the SD image, and
actual waveform (actual world signals), corresponding to the four
HD pixels from the left of the six continuous HD pixels in the X
direction contained in the region shown in FIG. 294.
FIG. 296 is a diagram illustrating a difference image of the
prediction image in FIG. 293 and an HD image.
FIG. 297 is a diagram plotting each of specific pixel values of the
HD image in FIG. 293, specific pixel values of the SD image, and
actual waveform (actual world signals), corresponding to the four
HD pixels from the left of the six continuous HD pixels in the X
direction contained in the region shown in FIG. 296.
FIG. 298 is a diagram for describing understanding obtained based
on the contents shown in FIG. 295 through FIG. 297.
FIG. 299 is a block diagram for describing a configuration example
of the class classification adaptation processing correction unit
of the image generating unit shown in FIG. 289.
FIG. 300 is a block diagram for describing a detailed configuration
example of the learning unit for the class classification
adaptation processing correction shown in FIG. 291.
FIG. 301 is a diagram for describing in-pixel gradient.
FIG. 302 is a diagram illustrating the SD image shown in FIG. 293,
and a features image having as the pixel value thereof the in-pixel
gradient of each of the pixels of the SD image.
FIG. 303 is a diagram for describing an in-pixel gradient
calculation method.
FIG. 304 is a diagram for describing an in-pixel gradient
calculation method.
FIG. 305 is a flowchart for describing the image generating
processing which the image generating unit of the configuration
shown in FIG. 289 executes.
FIG. 306 is a flowchart describing detailed input image class
classification adaptation processing in the image generating
processing in FIG. 305.
FIG. 307 is a flowchart for describing detailed correction
processing of the class classification adaptation processing in the
image generating processing in FIG. 305.
FIG. 308 is a diagram for describing an example of a class tap
array.
FIG. 309 is a diagram for describing an example of class
classification.
FIG. 310 is a diagram for describing an example of a prediction tap
array.
FIG. 311 is a flowchart for describing learning processing of the
learning device shown in FIG. 291.
FIG. 312 is a flowchart for describing detailed learning processing
for the class classification adaptation processing in the learning
processing shown in FIG. 311.
FIG. 313 is a flowchart for describing detailed learning processing
for the class classification adaptation processing correction in
the learning processing shown in FIG. 311.
FIG. 314 is a diagram illustrating the prediction image shown in
FIG. 293, and an image wherein a correction image is added to the
prediction image (the image generated by the image generating unit
shown in FIG. 289).
FIG. 315 is a block diagram describing a first configuration
example of a signal processing device using a hybrid technique,
which is another example of an embodiment of the signal processing
device shown in FIG. 1.
FIG. 316 is a block diagram for describing a configuration example
of an image generating unit for executing the class classification
adaptation processing of the signal processing device shown in FIG.
315.
FIG. 317 is a block diagram for describing a configuration example
of the learning device as to the image generating unit shown in
FIG. 316.
FIG. 318 is a flowchart for describing the processing of signals
executed by the signal processing device of the configuration shown
in FIG. 315.
FIG. 319 is a flowchart for describing the details of executing
processing of the class classification adaptation processing of the
signal processing in FIG. 318.
FIG. 320 is a flowchart for describing the learning processing of
the learning device shown in FIG. 317.
FIG. 321 is a block diagram describing a second configuration
example of a signal processing device using a hybrid technique,
which is another example of an embodiment of the signal processing
device shown in FIG. 1.
FIG. 322 is a flowchart for describing signal processing which the
signal processing device of the configuration shown in FIG. 319
executes.
FIG. 323 is a block diagram describing a third configuration
example of a signal processing device using a hybrid technique,
which is another example of an embodiment of the signal processing
device shown in FIG. 1.
FIG. 324 is a flowchart for describing signal processing which the
signal processing device of the configuration shown in FIG. 321
executes.
FIG. 325 is a block diagram describing a fourth configuration
example of a signal processing device using a hybrid technique,
which is another example of an embodiment of the signal processing
device shown in FIG. 1.
FIG. 326 is a flowchart for describing signal processing which the
signal processing device of the configuration shown in FIG. 323
executes.
FIG. 327 is a block diagram describing a fifth configuration
example of a signal processing device using a hybrid technique,
which is another example of an embodiment of the signal processing
device shown in FIG. 1.
FIG. 328 is a flowchart for describing signal processing which the
signal processing device of the configuration shown in FIG. 325
executes.
FIG. 329 is a block diagram illustrating the configuration of
another embodiment of the data continuity detecting unit.
FIG. 330 is a flowchart for describing data continuity detecting
processing with the data continuity detecting unit shown in FIG.
329.
FIG. 331 is a diagram describing the configuration of an optical
block.
FIG. 332 is a diagram describing the configuration of the optical
block.
FIG. 333 is a diagram describing the configuration of an OLPF.
FIG. 334 is a diagram describing the function of the OLPF.
FIG. 335 is a diagram describing the function of the OLPF.
FIG. 336 is a block diagram illustrating the other configuration of
the signal processing device according to the present
invention.
FIG. 337 is a block diagram illustrating the configuration of the
OLPF removal unit shown in FIG. 336.
FIG. 338 is a diagram illustrating an example of a class tap.
FIG. 339 is a flowchart for describing signal processing with the
signal processing device shown in FIG. 336.
FIG. 340 is a flowchart for describing OLPF removal processing,
which is the processing in step S5101 of the flowchart shown in
FIG. 339.
FIG. 341 is a learning device for learning the coefficient of the
OLPF removal unit shown in FIG. 337.
FIG. 342 is a diagram for describing a learning method.
FIG. 343 is a diagram for describing a tutor image and a student
image.
FIG. 344 is a block diagram illustrating the configurations of the
tutor image generating unit and student image generating unit of
the learning device shown in FIG. 342.
FIG. 345 is a diagram describing a method for generating a student
image and a tutor image.
FIG. 346 is a diagram for describing an OLPF simulation method.
FIG. 347 is a diagram illustrating an example of a tutor image.
FIG. 348 is a diagram illustrating an example of a student
image.
FIG. 349 is a flowchart for describing the processing for
learning.
FIG. 350 is a diagram illustrating an image subjected to the OLPF
removal processing.
FIG. 351 is a diagram for describing comparison between an image
subjected to the OLPF removal processing and an image not subjected
to the OLPF removal processing.
FIG. 352 is a block diagram illustrating the other configuration
example of the actual world estimating unit.
FIG. 353 is a diagram for describing influence by OLPF.
FIG. 354 is a diagram for describing influence by OLPF.
FIG. 355 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 352.
FIG. 356 is a diagram illustrating an example of a tap to be
extracted.
FIG. 357 is a diagram for comparing an image generated from the
approximation function of the actual world estimated by the actual
world estimating unit shown in FIG. 352 with an image generated
with a technique other than that.
FIG. 358 is a diagram for comparing an image generated from the
approximation function of the actual world estimated by the actual
world estimating unit shown in FIG. 352 with an image generated
with a technique other than that.
FIG. 359 is a block diagram illustrating the other configuration of
the signal processing device.
FIG. 360 is a flowchart for describing signal processing with the
signal processing device shown in FIG. 359.
FIG. 361 is a block diagram illustrating the configuration of a
learning device for learning the coefficient of the signal
processing device shown in FIG. 359.
FIG. 362 is a block diagram illustrating the configuration of the
tutor image generating unit and student image generating unit shown
in FIG. 361.
FIG. 363 is a flowchart for describing the processing of learning
with the learning device shown in FIG. 361.
FIG. 364 is a diagram for describing the relationship between
various types of image processing.
FIG. 365 is a diagram for describing actual world estimation with
an approximation function made up of a continuous function.
FIG. 366 is a diagram for describing an approximation function made
up of a discontinuous function.
FIG. 367 is a diagram for describing an approximation function made
up of a continuous function and a discontinuous function.
FIG. 368 is a diagram for describing a method for obtaining pixel
values using an approximation function made up of a discontinuous
function.
FIG. 369 is a block diagram illustrating the other configuration of
the actual world estimating unit.
FIG. 370 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 369.
FIG. 371 is a diagram illustrating an example of a tap to be
extracted.
FIG. 372 is a diagram for describing an approximation function made
up of a discontinuous function on the X-t plane.
FIG. 373 is a diagram for describing the other example of a tap to
be extracted.
FIG. 374 is a diagram for describing an approximation function made
up of a two-dimensional discontinuous function.
FIG. 375 is a diagram for describing an approximation function made
up of a two-dimensional discontinuous function.
FIG. 376 is a diagram for describing a volume rate for each pixel
of interest region.
FIG. 377 is a block diagram illustrating the other configuration of
the actual world estimating unit.
FIG. 378 is a flowchart for describing the processing of actual
world estimation with the actual world estimating unit shown in
FIG. 377.
FIG. 379 is a diagram for describing the other example of a tap to
be extracted.
FIG. 380 is a diagram for describing an approximation function made
up of a two-dimensional discontinuous function.
FIG. 381 is a diagram for describing an approximation function made
up of a two-dimensional discontinuous function.
FIG. 382 is a diagram for describing an approximation function made
up of a continuous function of a polynomial for each region.
FIG. 383 is a diagram for describing an approximation function made
up of a discontinuous function of a polynomial for each region.
FIG. 384 is a block diagram describing the other configuration of
the image generating unit.
FIG. 385 is a flowchart for describing the image generating
processing with the image generating unit shown in FIG. 384.
FIG. 386 is a diagram for describing a method for generating a
quadruple density pixel.
FIG. 387 is a diagram for describing the relationship between the
conventional technique and the case of employing an approximation
function made up of a discontinuous function.
FIG. 388 is a block diagram for describing the other configuration
of the image generating unit.
FIG. 389 is a flowchart for describing the image generating
processing with the image generating unit shown in FIG. 388.
FIG. 390 is a diagram for describing a pixel of interest.
FIG. 391 is a diagram for describing a method for computing the
pixel value of a pixel of interest.
FIG. 392 is a diagram for describing the processing result using an
approximation function made up of a discontinuous function in the
spatial directions and the other processing results.
FIG. 393 is a diagram for describing the processing result using an
approximation function made up of a discontinuous function and the
other processing results.
FIG. 394 is a diagram for describing imaging by a sensor.
FIG. 395 is a diagram describing the placement of pixels.
FIG. 396 is a diagram describing operation of detecting
devices.
FIG. 397 is a diagram for describing an image obtained by imaging
an object corresponding to the moving foreground, and an object
corresponding to the still background.
FIG. 398 is a diagram for describing a background region,
foreground region, mixed region, covered background region, and
uncovered background region.
FIG. 399 is a model diagram for expanding in the time direction the
pixel values of pixels adjacently arrayed in a row in an image on
which an object corresponding to the still foreground, and an
object corresponding to the still background are imaged.
FIG. 400 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 401 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 402 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 403 is a diagram illustrating an example wherein pixels
belonged to a foreground region, background region, and mixed
region are extracted.
FIG. 404 is a diagram illustrating correspondence with a model
wherein pixels and the pixel values thereof are expanded in the
time direction.
FIG. 405 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 406 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 407 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 408 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 409 is a model diagram wherein pixel values are expanded in
the time direction, and a period corresponding to shutter time is
divided.
FIG. 410 is a diagram for describing the processing result using an
approximation function made up of a discontinuous function in the
time-space directions and the other processing results.
FIG. 411 is a diagram for describing an image including movement
blurring in the horizontal direction.
FIG. 412 is a diagram for describing the processing result of the
image shown in FIG. 411 using an approximation function made up of
a discontinuous function in the time-space directions and the other
processing results.
FIG. 413 is a diagram for describing an image including movement
blurring in the oblique direction.
FIG. 414 is a diagram for describing the processing result of the
image shown in FIG. 413 using an approximation function made up of
a discontinuous function in the time-space directions and the other
processing results.
FIG. 415 is a diagram illustrating the processing result of an
image including movement blurring in the oblique direction using an
approximation function made up of a discontinuous function in the
time-space directions.
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 1 illustrates the principle of the present invention. As shown
in the drawing, events (phenomena) in an actual world 1 having
dimensions such as space, time, mass, and so forth, are acquired by
a sensor 2, and formed into data. Events in the actual world 1
refer to light (images), sound, pressure, temperature, mass,
humidity, brightness/darkness, or acts, and so forth. The events in
the actual world 1 are distributed in the space-time directions.
For example, an image of the actual world 1 is a distribution of
the intensity of light of the actual world 1 in the space-time
directions.
Taking note of the sensor 2, of the events in the actual world 1
having the dimensions of space, time, and mass, the events in the
actual world 1 which the sensor 2 can acquire, are converted into
data 3 by the sensor 2. It can be said that information indicating
events in the actual world 1 are acquired by the sensor 2.
That is to say, the sensor 2 converts information indicating events
in the actual world 1, into data 3. It can be said that signals
which are information indicating the events (phenomena) in the
actual world 1 having dimensions such as space, time, and mass, are
acquired by the sensor 2 and formed into data.
Hereafter, the distribution of events such as light (images),
sound, pressure, temperature, mass, humidity, rightness/darkness,
or smells, and so forth, in the actual world 1, will be referred to
as signals of the actual world 1, which are information indicating
events. Also, signals which are information indicating events of
the actual world 1 will also be referred to simply as signals of
the actual world 1. In the present Specification, signals are to be
understood to include phenomena and events, and also include those
wherein there is no intent on the transmitting side.
The data 3 (detected signals) output from the sensor 2 is
information obtained by projecting the information indicating the
events of the actual world 1 on a space-time having a lower
dimension than the actual world 1. For example, the data 3 which is
image data of a moving image, is information obtained by projecting
an image of the three-dimensional space direction and time
direction of the actual world 1 on the time-space having the
two-dimensional space direction and time direction. Also, in the
event that the data 3 is digital data for example, the data 3 is
rounded off according to the sampling increments. In the event that
the data 3 is analog data, information of the data 3 is either
compressed according to the dynamic range, or a part of the
information has been deleted by a limiter or the like.
Thus, by projecting the signals shown are information indicating
events in the actual world 1 having a predetermined number of
dimensions onto data 3 (detection signals), a part of the
information indicating events in the actual world 1 is dropped.
That is to say, a part of the information indicating events in the
actual world 1 is dropped from the data 3 which the sensor 2
outputs.
However, even though a part of the information indicating events in
the actual world 1 is dropped due to projection, the data 3
includes useful information for estimating the signals which are
information indicating events (phenomena) in the actual world
1.
With the present invention, information having continuity contained
in the data 3 is used as useful information for estimating the
signals which is information of the actual world 1. Continuity is a
concept which is newly defined.
Taking note of the actual world 1, events in the actual world 1
include characteristics which are constant in predetermined
dimensional directions. For example, an object (corporeal object)
in the actual world 1 either has shape, pattern, or color that is
continuous in the space direction or time direction, or has
repeated patterns of shape, pattern, or color.
Accordingly, the information indicating the events in actual world
1 includes characteristics constant in a predetermined dimensional
direction.
With a more specific example, a linear object such as a string,
cord, or rope, has a characteristic which is constant in the
length-wise direction, i.e., the spatial direction, that the
cross-sectional shape is the same at arbitrary positions in the
length-wise direction. The constant characteristic in the spatial
direction that the cross-sectional shape is the same at arbitrary
positions in the length-wise direction comes from the
characteristic that the linear object is long.
Accordingly, an image of the linear object has a characteristic
which is constant in the length-wise direction, i.e., the spatial
direction, that the cross-sectional shape is the same, at arbitrary
positions in the length-wise direction.
Also, a monotone object, which is a corporeal object, having an
expanse in the spatial direction, can be said to have a constant
characteristic of having the same color in the spatial direction
regardless of the part thereof.
In the same way, an image of a monotone object, which is a
corporeal object, having an expanse in the spatial direction, can
be said to have a constant characteristic of having the same color
in the spatial direction regardless of the part thereof.
In this way, events in the actual world 1 (real world) have
characteristics which are constant in predetermined dimensional
directions, so signals of the actual world 1 have characteristics
which are constant in predetermined dimensional directions.
In the present Specification, such characteristics which are
constant in predetermined dimensional directions will be called
continuity. Continuity of the signals of the actual world 1 (real
world) means the characteristics which are constant in
predetermined dimensional directions which the signals indicating
the events of the actual world 1 (real world) have.
Countless such continuities exist in the actual world 1 (real
world).
Next, taking note of the data 3, the data 3 is obtained by signals
which is information indicating events of the actual world 1 having
predetermined dimensions being projected by the sensor 2, and
includes continuity corresponding to the continuity of signals in
the real world. It can be said that the data 3 includes continuity
wherein the continuity of actual world signals has been
projected.
However, as described above, in the data 3 output from the sensor
2, a part of the information of the actual world 1 has been lost,
so a part of the continuity contained in the signals of the actual
world 1 (real world) is lost.
In other words., the data 3 contains a part of the continuity
within the continuity of the signals of the actual world 1 (real
world) as data continuity. Data continuity means characteristics
which are constant in predetermined dimensional directions, which
the data 3 has.
With the present invention, the data continuity which the data 3
has is used as significant data for estimating signals which are
information indicating events of the actual world 1.
For example, with the present invention, information indicating an
event in the actual world 1 which has been lost is generated by
signals processing of the data 3, using data continuity.
Now, with the present invention, of the length (space), time, and
mass, which are dimensions of signals serving as information
indicating events in the actual world 1, continuity in the spatial
direction or time direction, are used.
Returning to FIG. 1, the sensor 2 is formed of, for example, a
digital still camera, a video camera, or the like, and takes images
of the actual world 1, and outputs the image data which is the
obtained data 3, to a signal processing device 4. The sensor 2 may
also be a thermography device, a pressure sensor using
photo-elasticity, or the like.
The signal processing device 4 is configured of, for example, a
personal computer or the like.
The signal processing device 4 is configured as shown in FIG. 2,
for example. A CPU (Central Processing Unit) 21 executes various
types of processing following programs stored in ROM (Read Only
Memory) 22 or the storage unit 28. RAM (Random Access Memory) 23
stores programs to be executed by the CPU 21, data, and so forth,
as suitable. The CPU 21, ROM 22, and RAM 23, are mutually connected
by a bus 24.
Also connected to the CPU 21 is an input/output interface 25 via
the bus 24. An input device 26 made up of a keyboard, mouse,
microphone, and so forth, and an output unit 27 made up of a
display, speaker, and so forth, are connected to the input/output
interface 25. The CPU 21 executes various types of processing
corresponding to commands input from the input unit 26. The CPU 21
then outputs images and audio and the like obtained as a result of
processing to the output unit 27.
A storage unit 28 connected to the input/output interface 25 is
configured of a hard disk for example, and stores the programs and
various types of data which the CPU 21 executes. A communication
unit 29 communicates with external devices via the Internet and
other networks. In the case of this example, the communication unit
29 acts as an acquiring unit for capturing data 3 output from the
sensor 2.
Also, an arrangement may be made wherein programs are obtained via
the communication unit 29 and stored in the storage unit 28.
A drive 30 connected to the input/output interface 25 drives a
magnetic disk 51, optical disk 52, magneto-optical disk 53, or
semiconductor memory 54 or the like mounted thereto, and obtains
programs and data recorded therein. The obtained programs and data
are transferred to the storage unit 28 as necessary and stored.
FIG. 3 is a block diagram illustrating a signal processing device
4.
Note that whether the functions of the signal processing device 4
are realized by hardware or realized by software is irrelevant.
That is to say, the block diagrams in the present Specification may
be taken to be hardware block diagrams or may be taken to be
software function block diagrams.
With the signal processing device 4 shown in FIG. 3, image data
which is an example of the data 3 is input, and the continuity of
the data is detected from the input image data (input image). Next,
the signals of the actual world 1 acquired by the sensor 2 are
estimated from the continuity of the data detected. Then, based on
the estimated signals of the actual world 1, an image is generated,
and the generated image (output image) is output. That is to say,
FIG. 3 is a diagram illustrating the configuration of the signal
processing device 4 which is an image processing device.
The input image (image data which is an example of the data 3)
input to the signal processing device 4 is supplied to a data
continuity detecting unit 101 and actual world estimating unit
102.
The data continuity detecting unit 101 detects the continuity of
the data from the input image, and supplies data continuity
information indicating the detected continuity to the actual world
estimating unit 102 and an image generating unit 103. The data
continuity information includes, for example, the position of a
region of pixels having continuity of data, the direction of a
region of pixels having continuity of data (the angle or gradient
of the time direction and space direction), or the length of a
region of pixels having continuity of data, or the like in the
input image. Detailed configuration of the data continuity
detecting unit 101 will be described later.
The actual world estimating unit 102 estimates the signals of the
actual world 1, based on the input image and the data continuity
information supplied from the data continuity detecting unit 101.
That is to say, the actual world estimating unit 102 estimates an
image which is the signals of the actual world cast into the sensor
2 at the time that the input image was acquired. The actual world
estimating unit 102 supplies the actual world estimation
information indicating the results of the estimation of the signals
of the actual world 1, to the image generating unit 103. The
detailed configuration of the actual world estimating unit 102 will
be described later.
The image generating unit 103 generates signals further
approximating the signals of the actual world 1, based on the
actual world estimation information indicating the estimated
signals of the actual world 1, supplied from the actual world
estimating unit 102, and outputs the generated signals. Or, the
image generating unit 103 generates signals further approximating
the signals of the actual world 1, based on the data continuity
information supplied from the data continuity detecting unit 101,
and the actual world estimation information indicating the
estimated signals of the actual world 1, supplied from the actual
world estimating unit 102, and outputs the generated signals.
That is to say, the image generating unit 103 generates an image
further approximating the image of the actual world 1 based on the
actual world estimation information, and outputs the generated
image as an output image. Or, the image generating unit 103
generates an image further approximating the image of the actual
world 1 based on the data continuity information and actual world
estimation information, and outputs the generated image as an
output image.
For example, the image generating unit 103 generates an image with
higher resolution in the spatial direction or time direction in
comparison with the input image, by integrating the estimated image
of the actual world 1 within a desired range of the spatial
direction or time direction, based on the actual world estimation
information, and outputs the generated image as an output image.
For example, the image generating unit 103 generates an image by
extrapolation/interpolation, and outputs the generated image as an
output image.
Detailed configuration of the image generating unit 103 will be
described later.
Next, the principle of the present invention will be described with
reference to FIG. 4 through FIG. 7.
FIG. 4 is a diagram describing the principle of processing with a
conventional signal processing device 121. The conventional signal
processing device 121 takes the data 3 as the reference for
processing, and executes processing such as increasing resolution
and the like with the data 3 as the object of processing. With the
conventional signal processing device 121, the actual world 1 is
never taken into consideration, and the data 3 is the ultimate
reference, so information exceeding the information contained in
the data 3 can not be obtained as output.
Also, with the conventional signal processing device 121,
distortion in the data 3 due to the sensor 2 (difference between
the signals which are information of the actual world 1, and the
data 3) is not taken into consideration whatsoever, so the
conventional signal processing device 121 outputs signals still
containing the distortion. Further, depending on the processing
performed by the signal processing device 121, the distortion due
to the sensor 2 present within the data 3 is further amplified, and
data containing the amplified distortion is output.
Thus, with conventional signals processing, (the signals of) the
actual world 1, from which the data 3 has been obtained, was never
taken into consideration. In other words, with the conventional
signal processing, the actual world 1 was understood within the
framework of the information contained in the data 3, so the limits
of the signal processing are determined by the information and
distortion contained in the data 3. The present Applicant has
separately proposed signal processing taking into consideration the
actual world 1, but this did not take into consideration the
later-described continuity.
In contrast with this, with the signal processing according to the
present invention, processing is executed taking (the signals of)
the actual world 1 into consideration in an explicit manner.
FIG. 5 is a diagram for describing the principle of the processing
at the signal processing device 4 according to the present
invention.
This is the same as the conventional arrangement wherein signals,
which are information indicating events of the actual world 1, are
obtained by the sensor 2, and the sensor 2 outputs data 3 wherein
the signals which are information of the actual world 1 are
projected.
However, with the present invention, signals, which are information
indicating events of the actual world 1, obtained by the sensor 2,
are explicitly taken into consideration. That is to say, signal
processing is performed conscious of the fact that the data 3
contains distortion due to the sensor 2 (difference between the
signals which are information of the actual world 1, and the data
3).
Thus, with the signal processing according to the present
invention, the processing results are not restricted due to the
information contained in the data 3 and the distortion, and for
example, processing results which are more accurate and which have
higher precision than conventionally can be obtained with regard to
events in the actual world 1. That is to say, with the present
invention, processing results which are more accurate and which
have higher precision can be obtained with regard to signals, which
are information indicating events of the actual world 1, input to
the sensor 2.
FIG. 6 and FIG. 7 are diagrams for describing the principle of the
present invention in greater detail.
As shown in FIG. 6, signals of the actual world, which are an image
for example, are image on the photoreception face of a CCD (Charge
Coupled Device) which is an example of the sensor 2, by an optical
system 141 made up of lenses, an optical LPF (Low Pass Filter), and
the like. The CCD, which is an example of the sensor 2, has
integration properties, so difference is generated in the data 3
output from the CCD as to the image of the actual world 1. Details
of the integration properties of the sensor 2 will be described
later.
With the signal processing according to the present invention, the
relationship between the image of the actual world 1 obtained by
the CCD, and the data 3 taken by the CCD and output, is explicitly
taken into consideration. That is to say, the relationship between
the data 3 and the signals which is information of the actual world
obtained by the sensor 2, is explicitly taken into
consideration.
More specifically, as shown in FIG. 7, the signal processing device
4 uses a model 161 to approximate (describe) the actual world 1.
The model 161 is represented by, for example, N variables. More
accurately, the model 161 approximates (describes) signals of the
actual world 1.
In order to predict the model 161, the signal processing device 4
extracts M pieces of data 162 from the data 3. At the time of
extracting the M pieces of data 162 from the data 3, the signal
processing device 4 uses the continuity of the data contained in
the data 3. In other words, the signal processing device 4 extracts
data 162 for predicting the model 161, based o the continuity of
the data contained in the data 3. Consequently, the model 161 is
constrained by the continuity of the data.
That is to say, the model 161 approximates (information (signals)
indicating) events of the actual world having continuity (constant
characteristics in a predetermined dimensional direction), which
generates the data continuity in the data 3.
Now, in the event that the number M of the data 162 is N or more,
which is the number of variables of the model, the model 161
represented by the N variables can be predicted, from the M pieces
of the data 162.
In this way, the signal processing device 4 can take into
consideration the signals which are information of the actual world
1, by predicting the model 161 approximating (describing) the
(signals of the) actual world 1.
Next, the integration effects of the sensor 2 will be
described.
An image sensor such as a CCD or CMOS (Complementary Metal-Oxide
Semiconductor), which is the sensor 2 for taking images, projects
signals, which are information of the real world, onto
two-dimensional data, at the time of imaging the real world. The
pixels of the image sensor each have a predetermined area, as a
so-called photoreception face (photoreception region). Incident
light to the photoreception face having a predetermined area is
integrated in the space direction and time direction for each
pixel, and is converted into a single pixel value for each
pixel.
The space-time integration of images will be described with
reference to FIG. 8 through FIG. 11.
An image sensor images a subject (object) in the real world, and
outputs the obtained image data as a result of imagining in
increments of single frames. That is to say, the image sensor
acquires signals of the actual world 1 which is light reflected off
of the subject of the actual world 1, and outputs the data 3.
For example, the image sensor outputs image data of 30 frames per
second. In this case, the exposure time of the image sensor can be
made to be 1/30 seconds. The exposure time is the time from the
image sensor starting conversion of incident light into electric
charge, to ending of the conversion of incident light into electric
charge. Hereafter, the exposure time will also be called shutter
time.
FIG. 8 is a diagram describing an example of a pixel array on the
image sensor. In FIG. 8, A through I denote individual pixels. The
pixels are placed on a plane corresponding to the image displayed
by the image data. A single detecting element corresponding to a
single pixel is placed on the image sensor. At the time of the
image sensor taking images of the actual world 1, the one detecting
element outputs one pixel value corresponding to the one pixel
making up the image data. For example, the position in the spatial
direction X (X coordinate) of the detecting element corresponds to
the horizontal position on the image displayed by the image data,
and the position in the spatial direction Y (Y coordinate) of the
detecting element corresponds to the vertical position on the image
displayed by the image data.
Distribution of intensity of light of the actual world 1 has
expanse in the three-dimensional spatial directions and the time
direction, but the image sensor acquires light of the actual world
1 in two-dimensional spatial directions and the time direction, and
generates data 3 representing the distribution of intensity of
light in the two-dimensional spatial directions and the time
direction.
As shown in FIG. 9, the detecting device which is a CCD for
example, converts light cast onto the photoreception face
(photoreception region) (detecting region) into electric charge
during a period corresponding to the shutter time, and accumulates
the converted charge. The light is information (signals) of the
actual world 1 regarding which the intensity is determined by the
three-dimensional spatial position and point-in-time. The
distribution of intensity of light of the actual world 1 can be
represented by a function F(x, y, z, t), wherein position x, y, z,
in three-dimensional space, and point-in-time t, are variables.
The amount of charge accumulated in the detecting device which is a
CCD is approximately proportionate to the intensity of the light
cast onto the entire photoreception face having two-dimensional
spatial expanse, and the amount of time that light is cast
thereupon. The detecting device adds the charge converted from the
light cast onto the entire photoreception face, to the charge
already accumulated during a period corresponding to the shutter
time. That is to say, the detecting device integrates the light
cast onto the entire photoreception face having a two-dimensional
spatial expanse, and accumulates a change of an amount
corresponding to the integrated light during a period corresponding
to the shutter time. The detecting device can also be said to have
an integration effect regarding space (photoreception face) and
time (shutter time).
The charge accumulated in the detecting device is converted into a
voltage value by an unshown circuit, the voltage value is further
converted into a pixel value such as digital data or the like, and
is output as data 3. Accordingly, the individual pixel values
output from the image sensor have a value projected on
one-dimensional space, which is the result of integrating the
portion of the information (signals) of the actual world 1 having
time-space expanse with regard to the time direction of the shutter
time and the spatial direction of the photoreception face of the
detecting device.
That is to say, the pixel value of one pixel is represented as the
integration of F(x, y, t). F(x, y, t) is a function representing
the distribution of light intensity on the photoreception face of
the detecting device. For example, the pixel value P is represented
by Expression (1).
.intg..times..intg..times..intg..times..function..times.d.times.d.times.d
##EQU00001##
In Expression (1), x.sub.1 represents the spatial coordinate at the
left-side boundary of the photoreception face of the detecting
device (X coordinate). x.sub.2 represents the spatial coordinate at
the right-side boundary of the photoreception face of the detecting
device (X coordinate). In Expression (1), y.sub.1 represents the
spatial coordinate at the top-side boundary of the photoreception
face of the detecting device (Y coordinate). y.sub.2 represents the
spatial coordinate at the bottom-side boundary of the
photoreception face of the detecting device (Y coordinate). Also,
t.sub.1 represents the point-in-time at which conversion of
incident light into an electric charge was started. t.sub.2
represents the point-in-time at which conversion of incident light
into an electric charge was ended.
Note that actually, the gain of the pixel values of the image data
output from the image sensor is corrected for the overall
frame.
Each of the pixel values of the image data are integration values
of the light cast on the photoreception face of each of the
detecting elements of the image sensor, and of the light cast onto
the image sensor, waveforms of light of the actual world 1 finer
than the photoreception face of the detecting element are hidden in
the pixel value as integrated values.
Hereafter, in the present Specification, the waveform of signals
represented with a predetermined dimension as a reference may be
referred to simply as waveforms.
Thus, the image of the actual world 1 is integrated in the spatial
direction and time direction in increments of pixels, so a part of
the continuity of the image of the actual world 1 drops out from
the image data, so only another part of the continuity of the image
of the actual world 1 is left in the image data. Or, there may be
cases wherein continuity which has changed from the continuity of
the image of the actual world 1 is included in the image data.
Further description will be made regarding the integration effect
in the spatial direction for an image taken by an image sensor
having integration effects.
FIG. 10 is a diagram describing the relationship between incident
light to the detecting elements corresponding to the pixel D
through pixel F, and the pixel values. F(x) in FIG. 10 is an
example of a function representing the distribution of light
intensity of the actual world 1, having the coordinate x in the
spatial direction X in space (on the detecting device) as a
variable. In other words, F(x) is an example of a function
representing the distribution of light intensity of the actual
world 1, with the spatial direction Y and time direction constant.
In FIG. 10, L indicates the length in the spatial direction X of
the photoreception face of the detecting device corresponding to
the pixel D through pixel F.
The pixel value of a single pixel is represented as the integral of
F(x). For example, the pixel value P of the pixel E is represented
by Expression (2).
.intg..times..function..times.d ##EQU00002##
In the Expression (2), x.sub.1 represents the spatial coordinate in
the spatial direction X at the left-side boundary of the
photoreception face of the detecting device corresponding to the
pixel E. x.sub.2 represents the spatial coordinate in the spatial
direction X at the right-side boundary of the photoreception face
of the detecting device corresponding to the pixel E.
In the same way, further description will be made regarding the
integration effect in the time direction for an image taken by an
image sensor having integration effects.
FIG. 11 is a diagram for describing the relationship between time
elapsed, the incident light to a detecting element corresponding to
a single pixel, and the pixel value. F(t) in FIG. 11 is a function
representing the distribution of light intensity of the actual
world 1, having the point-in-time t as a variable. In other words,
F(t) is an example of a function representing the distribution of
light intensity of the actual world 1, with the spatial direction Y
and the spatial direction X constant. T.sub.s represents the
shutter time.
The frame #n-1 is a frame which is previous to the frame #n
time-wise, and the frame #n+1 is a frame following the frame #n
time-wise. That is to say, the frame #n-1, frame #n, and frame
#n+1, are displayed in the order of frame #n-1, frame #n, and frame
#n+1.
Note that in the example shown in FIG. 11, the shutter time t.sub.s
and the frame intervals are the same.
The pixel value of a single pixel is represented as the integral of
F(x). For example, the pixel value P of the pixel of frame #n for
example, is represented by Expression (3).
.intg..times..function..times.d ##EQU00003##
In the Expression (3), t.sub.1 represents the time at which
conversion of incident light into an electric charge was started.
t.sub.2 represents the time at which conversion of incident light
into an electric charge was ended.
Hereafter, the integration effect in the spatial direction by the
sensor 2 will be referred to simply as spatial integration effect,
and the integration effect in the time direction by the sensor 2
also will be referred to simply as time integration effect. Also,
space integration effects or time integration effects will be
simply called integration effects.
Next, description will be made regarding an example of continuity
of data included in the data 3 acquired by the image sensor having
integration effects.
FIG. 12 is a diagram illustrating a linear object of the actual
world 1 (e.g., a fine line), i.e., an example of distribution of
light intensity. In FIG. 12, the position to the upper side of the
drawing indicates the intensity (level) of light, the position to
the upper right side of the drawing indicates the position in the
spatial direction X which is one direction of the spatial
directions of the image, and the position to the right side of the
drawing indicates the position in the spatial direction Y which is
the other direction of the spatial directions of the image.
The image of the linear object of the actual world 1 includes
predetermined continuity. That is to say, the image shown in FIG.
12 has continuity in that the cross-sectional shape (the change in
level as to the change in position in the direction orthogonal to
the length direction), at any arbitrary position in the length
direction.
FIG. 13 is a diagram illustrating an example of pixel values of
image data obtained by actual image-taking, corresponding to the
image shown in FIG. 12.
FIG. 14 is a model diagram of the image data shown in FIG. 13.
The model diagram shown in FIG. 14 is a model diagram of image data
obtained by imaging, with the image sensor, an image of a linear
object having a diameter shorter than the length L of the
photoreception face of each pixel, and extending in a direction
offset from the array of the pixels of the image sensor (the
vertical or horizontal array of the pixels). The image cast into
the image sensor at the time that the image data shown in FIG. 14
was acquired is an image of the linear object of the actual world 1
shown in FIG. 12.
In FIG. 14, the position to the upper side of the drawing indicates
the pixel value, the position to the upper right side of the
drawing indicates the position in the spatial direction X which is
one direction of the spatial directions of the image, and the
position to the right side of the drawing indicates the position in
the spatial direction Y which is the other direction of the spatial
directions of the image. The direction indicating the pixel value
in FIG. 14 corresponds to the direction of level in FIG. 12, and
the spatial direction X and spatial direction Y in FIG. 14 also are
the same as the directions in FIG. 12.
In the event of taking an image of a linear object having a
diameter narrower than the length L of the photoreception face of
each pixel with the image sensor, the linear object is represented
in the image data obtained as a result of the image-taking as
multiple arc shapes (half-discs) having a predetermined length
which are arrayed in a diagonally-offset fashion, in a model
representation, for example. The arc shapes are of approximately
the same shape. One arc shape is formed on one row of pixels
vertically, or is formed on one row of pixels horizontally. For
example, one arc shape shown in FIG. 14 is formed on one row of
pixels vertically.
Thus, with the image data taken and obtained by the image sensor
for example, the continuity in that the cross-sectional shape in
the spatial direction Y at any arbitrary position in the length
direction which the linear object image of the actual world 1 had,
is lost. Also, it can be said that the continuity, which the linear
object image of the actual world 1 had, has changed into continuity
in that arc shapes of the same shape formed on one row of pixels
vertically or formed on one row of pixels horizontally are arrayed
at predetermined intervals.
FIG. 15 is a diagram illustrating an image in the actual world 1 of
an object having a straight edge, and is of a monotone color
different from that of the background, i.e., an example of
distribution of light intensity. In FIG. 15, the position to the
upper side of the drawing indicates the intensity (level) of light,
the position to the upper right side of the drawing indicates the
position in the spatial direction X which is one direction of the
spatial directions of the image, and the position to the right side
of the drawing indicates the position in the spatial direction Y
which is the other direction of the spatial directions of the
image.
The image of the object of the actual world 1 which has a straight
edge and is of a monotone color different from that of the
background, includes predetermined continuity. That is to say, the
image shown in FIG. 15 has continuity in that the cross-sectional
shape (the change in level as to the change in position in the
direction orthogonal to the length direction) is the same at any
arbitrary position in the length direction.
FIG. 16 is a diagram illustrating an example of pixel values of the
image data obtained by actual image-taking, corresponding to the
image shown in FIG. 15. As shown in FIG. 16, the image data is in a
stepped shape, since the image data is made up of pixel values in
increments of pixels.
FIG. 17 is a model diagram illustrating the image data shown in
FIG. 16.
The model diagram shown in FIG. 17 is a model diagram of image data
obtained by taking, with the image sensor, an image of the object
of the actual world 1 which has a straight edge and is of a
monotone color different from that of the background, and extending
in a direction offset from the array of the pixels of the image
sensor (the vertical or horizontal array of the pixels). The image
cast into the image sensor at the time that the image data shown in
FIG. 17 was acquired is an image of the object of the actual world
1 which has a straight edge and is of a monotone color different
from that of the background, shown in FIG. 15.
In FIG. 17, the position to the upper side of the drawing indicates
the pixel value, the position to the upper right side of the
drawing indicates the position in the spatial direction X which is
one direction of the spatial directions of the image, and the
position to the right side of the drawing indicates the position in
the spatial direction Y which is the other direction of the spatial
directions of the image. The direction indicating the pixel value
in FIG. 17 corresponds to the direction of level in FIG. 15, and
the spatial direction X and spatial direction Y in FIG. 17 also are
the same as the directions in FIG. 15.
In the event of taking an image of an object of the actual world 1
which has a straight edge and is of a monotone color different from
that of the background with an image sensor, the straight edge is
represented in the image data obtained as a result of the
image-taking as multiple pawl shapes having a predetermined length
which are arrayed in a diagonally-offset fashion, in a model
representation, for example. The pawl shapes are of approximately
the same shape. One pawl shape is formed on one row of pixels
vertically, or is formed on one row of pixels horizontally. For
example, one pawl shape shown in FIG. 17 is formed on one row of
pixels vertically.
Thus, the continuity of image of the object of the actual world 1
which has a straight edge and is of a monotone color different from
that of the background, in that the cross-sectional shape is the
same at any arbitrary position in the length direction of the edge,
for example, is lost in the image data obtained by imaging with an
image sensor. Also, it can be said that the continuity, which the
image of the object of the actual world 1 which has a straight edge
and is of a monotone color different from that of the background
had, has changed into continuity in that pawl shapes of the same
shape formed on one row of pixels vertically or formed on one row
of pixels horizontally are arrayed at predetermined intervals.
The data continuity detecting unit 101 detects such data continuity
of the data 3 which is an input image, for example. For example,
the data continuity detecting unit 101 detects data continuity by
detecting regions having a constant characteristic in a
predetermined dimensional direction. For example, the data
continuity detecting unit 101 detects a region wherein the same arc
shapes are arrayed at constant intervals, such as shown in FIG. 14.
Also, the data continuity detecting unit 101 detects a region
wherein the same pawl shapes are arrayed at constant intervals,
such as shown in FIG. 17.
Also, the data continuity detecting unit 101 detects continuity of
the data by detecting angle (gradient) in the spatial direction,
indicating an array of the same shapes.
Also, for example, the data continuity detecting unit 101 detects
continuity of data by detecting angle (movement) in the space
direction and time direction, indicating the array of the same
shapes in the space direction and the time direction.
Further, for example, the data continuity detecting unit 101
detects continuity in the data by detecting the length of the
region having constant characteristics in a predetermined
dimensional direction.
Hereafter, the portion of data 3 where the sensor 2 has projected
the image of the object of the actual world 1 which has a straight
edge and is of a monotone color different from that of the
background, will also be called a two-valued edge.
Next, the principle of the present invention will be described in
further detail.
As shown in FIG. 18, with conventional signal processing, desired
high-resolution data 181, for example, is generated from the data
3.
Conversely, with the signal processing according to the present
invention, the actual world 1 is estimated from the data 3, and the
high-resolution data 181 is generated based on the estimation
results. That is to say, as shown in FIG. 19, the actual world 1 is
estimated from the data 3, and the high-resolution data 181 is
generated based on the estimated actual world 1, taking into
consideration the data 3.
In order to generate the high-resolution data 181 from the actual
world 1, there is the need to take into consideration the
relationship between the actual world 1 and the data 3. For
example, how the actual world 1 is projected on the data 3 by the
sensor 2 which is a CCD, is taken into consideration.
The sensor 2 which is a CCD has integration properties as described
above. That is to say, one unit of the data 3 (e.g., pixel value)
can be calculated by integrating a signal of the actual world 1
with a detection region (e.g., photoreception face) of a detection
device (e.g., CCD) of the sensor 2.
Applying this to the high-resolution data 181, the high-resolution
data 181 can be obtained by applying processing, wherein a virtual
high-resolution sensor projects signals of the actual world 1 to
the data 3, to the estimated actual world 1.
In other words, as shown in FIG. 20, if the signals of the actual
world 1 can be estimated from the data 3, one value contained in
the high-resolution data 181 can be obtained by integrating signals
of the actual world 1 for each detection region of the detecting
elements of the virtual high-resolution sensor (in the time-space
direction).
For example, in the event that the change in signals of the actual
world 1 are smaller than the size of the detection region of the
detecting elements of the sensor 2, the data 3 cannot expresses the
small changes in the signals of the actual world 1. Accordingly,
high-resolution data 181 indicating small change of the signals of
the actual world 1 can be obtained by integrating the signals of
the actual world 1 estimated from the data 3 with each region (in
the time-space direction) that is smaller in comparison with the
change in signals of the actual world 1.
That is to say, integrating the signals of the estimated actual
world 1 with the detection region with regard to each detecting
element of the virtual high-resolution sensor enables the
high-resolution data 181 to be obtained.
With the present invention, the image generating unit 103 generates
the high-resolution data 181 by integrating the signals of the
estimated actual world 1 in the time-space direction regions of the
detecting elements of the virtual high-resolution sensor.
Next, with the present invention, in order to estimate the actual
world 1 from the data 3, the relationship between the data 3 and
the actual world 1, continuity, and a space mixture in the data 3,
are used.
Here, a mixture means a value in the data 3 wherein the signals of
two objects in the actual world 1 are mixed to yield a single
value.
A space mixture means the mixture of the signals of two objects in
the spatial direction due to the spatial integration effects of the
sensor 2.
The actual world 1 itself is made up of countless events, and
accordingly, in order to represent the actual world 1 itself with
mathematical expressions, for example, there is the need to have an
infinite number of variables. It is impossible to predict all
events of the actual world 1 from the data 3.
In the same way, it is impossible to predict all of the signals of
the actual world 1 from the data 3.
Accordingly, as shown in FIG. 21, with the present embodiment, of
the signals of the actual world 1, a portion which has continuity
and which can be expressed by the function f(x, y, z, t) is taken
note of, and the portion of the signals of the actual world 1 which
can be represented by the function f(x, y, z, t) and has continuity
is approximated with a model 161 represented by N variables. As
shown in FIG. 22, the model 161 is predicted from the M pieces of
data 162 in the data 3.
In order to enable the model 161 to be predicted from the M pieces
of data 162, first, there is the need to represent the model 161
with N variables based on the continuity, and second, to generate
an expression using the N variables which indicates the
relationship between the model 161 represented by the N variables
and the M pieces of data 162 based on the integral properties of
the sensor 2. Since the model 161 is represented by the N
variables, based on the continuity, it can be said that the
expression using the N variables that indicates the relationship
between the model 161 represented by the N variables and the M
pieces of data 162, describes the relationship between the part of
the signals of the actual world 1 having continuity, and the part
of the data 3 having data continuity.
In other words, the part of the signals of the actual world 1
having continuity, that is approximated by the model 161
represented by the N variables, generates data continuity in the
data 3.
The data continuity detecting unit 101 detects the part of the data
3 where data continuity has been generated by the part of the
signals of the actual world 1 having continuity, and the
characteristics of the part where data continuity has been
generated.
For example, as shown in FIG. 23, in an image of the object of the
actual world 1 which has a straight edge and is of a monotone color
different from that of the background, the edge at the position of
interest indicated by A in FIG. 23, has a gradient. The arrow B in
FIG. 23 indicates the gradient of the edge. A predetermined edge
gradient can be represented as an angle as to a reference axis or
as a direction as to a reference position. For example, a
predetermined edge gradient can be represented as the angle between
the coordinates axis of the spatial direction X and the edge. For
example, the predetermined edge gradient can be represented as the
direction indicated by the length of the spatial direction X and
the length of the spatial direction Y.
At the time that the image of the object of the actual world 1
which has a straight edge and is of a monotone color different from
that of the background is obtained at the sensor 2 and the data 3
is output, pawl shapes corresponding to the edge are arrayed in the
data 3 at the position corresponding to the position of interest
(A) of the edge in the image of the actual world 1, which is
indicated by A' in FIG. 23, and pawl shapes corresponding to the
edge are arrayed in the direction corresponding to the gradient of
the edge of the image in the actual world 1, in the direction of
the gradient indicated by B' in FIG. 23.
The model 161 represented with the N variables approximates such a
portion of the signals of the actual world 1 generating data
continuity in the data 3.
At the time of formulating an expression using the N variables
indicating the relationship between the model 161 represented with
the N variables and the M pieces of data 162, the part where data
continuity is generated in the data 3 is used.
In this case, in the data 3 shown in FIG. 24, taking note of the
values where data continuity is generated and which belong to a
mixed region, an expression is formulated with a value integrating
the signals of the actual world 1 as being equal to a value output
by the detecting element of the sensor 2. For example, multiple
expressions can be formulated regarding the multiple values in the
data 3 where data continuity is generated.
In FIG. 24, A denotes the position of interest of the edge, and A'
denotes (the position of) the pixel corresponding to the position
(A) of interest of the edge in the image of the actual world 1.
Now, a mixed region means a region of data in the data 3 wherein
the signals for two objects in the actual world 1 are mixed and
become one value. For example, a pixel value wherein, in the image
of the object of the actual world 1 which has a straight edge and
is of a monotone color different from that of the background in the
data 3, the image of the object having the straight edge and the
image of the background are integrated, belongs to a mixed
region.
FIG. 25 is a diagram illustrating signals for two objects in the
actual world 1 and values belonging to a mixed region, in a case of
formulating an expression.
FIG. 25 illustrates, to the left, signals of the actual world 1
corresponding to two objects in the actual world 1 having a
predetermined expansion in the spatial direction X and the spatial
direction Y, which are acquired at the detection region of a single
detecting element of the sensor 2. FIG. 25 illustrates, to the
right, a pixel value P of a single pixel in the data 3 wherein the
signals of the actual world 1 illustrated to the left in FIG. 25
have been projected by a single detecting element of the sensor 2.
That is to say, illustrates a pixel value P of a single pixel in
the data 3 wherein the signals of the actual world 1 corresponding
to two objects in the actual world 1 having a predetermined
expansion in the spatial direction X and the spatial direction Y
which are acquired by a single detecting element of the sensor 2,
have been projected.
L in FIG. 25 represents the level of the signal of the actual world
1 which is shown in white in FIG. 25, corresponding to one object
in the actual world 1. R in FIG. 25 represents the level of the
signal of the actual world 1 which is shown hatched in FIG. 25,
corresponding to the other object in the actual world 1.
Here, the mixture ratio .alpha. is the ratio of (the area of) the
signals corresponding to the two objects cast into the detecting
region of the one detecting element of the sensor 2 having a
predetermined expansion in the spatial direction X and the spatial
direction Y. For example, the mixture ratio .alpha. represents the
ratio of area of the level L signals cast into the detecting region
of the one detecting element of the sensor 2 having a predetermined
expansion in the spatial direction X and the spatial direction Y,
as to the area of the detecting region of a single detecting
element of the sensor 2.
In this case, the relationship between the level L, level R, and
the pixel value P, can be represented by Expression (4).
.alpha..times.L+(1-.alpha.).times.R=P (4)
Note that there may be cases wherein the level R may be taken as
the pixel value of the pixel in the data 3 positioned to the right
side of the pixel of interest, and there may be cases wherein the
level L may be taken as the pixel value of the pixel in the data 3
positioned to the left side of the pixel of interest.
Also, the time direction can be taken into consideration in the
same way as with the spatial direction for the mixture ratio
.alpha. and the mixed region. For example, in the event that an
object in the actual world 1 which is the object of image-taking,
is moving as to the sensor 2, the ratio of signals for the two
objects cast into the detecting region of the single detecting
element of the sensor 2 changes in the time direction. The signals
for the two objects regarding which the ratio changes in the time
direction, that have been cast into the detecting region of the
single detecting element of the sensor 2, are projected into a
single value of the data 3 by the detecting element of the sensor
2.
The mixture of signals for two objects in the time direction due to
time integration effects of the sensor 2 will be called time
mixture.
The data continuity detecting unit 101 detects regions of pixels in
the data 3 where signals of the actual world 1 for two objects in
the actual world 1, for example, have been projected. The data
continuity detecting unit 101 detects gradient in the data 3
corresponding to the gradient of an edge of an image in the actual
world 1, for example.
The actual world estimating unit 102 estimates the signals of the
actual world by formulating an expression using N variables,
representing the relationship between the model 161 represented by
the N variables and the M pieces of data 162, based on the region
of the pixels having a predetermined mixture ratio .alpha. detected
by the data continuity detecting unit 101 and the gradient of the
region, for example, and solving the formulated expression.
Description will be made further regarding specific estimation of
the actual world 1.
Of the signals of the actual world represented by the function F(x,
y, z, t) let us consider approximating the signals of the actual
world represented by the function F(x, y, t) at the cross-section
in the spatial direction Z (the position of the sensor 2), with an
approximation function f(x, y, t) determined by a position x in the
spatial direction X, a position y in the spatial direction Y, and a
point-in-time t.
Now, the detection region of the sensor 2 has an expanse in the
spatial direction X and the spatial direction Y. In other words,
the approximation function f(x, y, t) is a function approximating
the signals of the actual world 1 having an expanse in the spatial
direction and time direction, which are acquired with the sensor
2.
Let us say that projection of the signals of the actual world 1
yields a value P(x, y, t) of the data 3. The value P(x, y, t) of
the data 3 is a pixel value which the sensor 2 which is an image
sensor outputs, for example.
Now, in the event that the projection by the sensor 2 can be
formulated, the value obtained by projecting the approximation
function f(x, y, t) can be represented as a projection function
S(x, y, t).
Obtaining the projection function S(x, y, t) has the following
problems.
First, generally, the function F(x, y, z, t) representing the
signals of the actual world 1 can be a function with an infinite
number of orders.
Second, even if the signals of the actual world could be described
as a function, the projection function S(x, y, t) via projection of
the sensor 2 generally cannot be determined. That is to say, the
action of projection by the sensor 2, in other words, the
relationship between the input signals and output signals of the
sensor 2, is unknown, so the projection function S(x, y, t) cannot
be determined.
With regard to the first problem, let us consider expressing the
function f(x, y, t) approximating signals of the actual world 1
with the sum of products of the function f.sub.i(x, y, t) which is
a describable function (e.g., a function with a finite number of
orders) and variables w.sub.i.
Also, with regard to the second problem, formulating projection by
the sensor 2 allows us to describe the function S.sub.i(x, y, t)
from the description of the function f.sub.i(x, y, t).
That is to say, representing the function f(x, y, t) approximating
signals of the actual world 1 with the sum of products of the
function f.sub.i(x, y, t) and variables w.sub.i, the Expression (5)
can be obtained.
.function..times..times..function. ##EQU00004##
For example, as indicated in Expression (6), the relationship
between the data 3 and the signals of the actual world can be
formulated as shown in Expression (7) from Expression (5) by
formulating the projection of the sensor 2.
.function..intg..intg..intg..function..times.d.times.d.times.d.function..-
times..times..function. ##EQU00005##
In Expression (7), j represents the index of the data.
In the event that M data groups (j=1 through M) common with the N
variables w.sub.i (i=1 through N) exists in Expression (7),
Expression (8) is satisfied, so the model 161 of the actual world
can be obtained from data 3. N.ltoreq.M (8)
N is the number of variables representing the model 161
approximating the actual world 1. M is the number of pieces of data
162 include in the data 3.
Representing the function f(x, y, t) approximating the actual world
1 with Expression (5) allows the variable portion w.sub.i to be
handled independently. At this time, i represents the number of
variables. Also, the form of the function represented by f.sub.i
can be handed independently, and a desired function can be used for
f.sub.i.
Accordingly, the number N of the variables w.sub.i can be defined
without dependence on the function f.sub.i, and the variables
w.sub.i can be obtained from the relationship between the number N
of the variables w.sub.i and the number of pieces of data M.
That is to say, using the following three allows the actual world 1
to be estimated from the data 3.
First, the N variables are determined. That is to say, Expression
(5) is determined. This enables describing the actual world 1 using
continuity. For example, the signals of the actual world 1 can be
described with a model 161 wherein a cross-section is expressed
with a polynomial, and the same cross-sectional shape continues in
a constant direction.
Second, for example, projection by the sensor 2 is formulated,
describing Expression (7). For example, this is formulated such
that the results of integration of the signals of the actual world
2 are data 3.
Third, M pieces of data 162 are collected to satisfy Expression
(8). For example, the data 162 is collected from a region having
data continuity that has been detected with the data continuity
detecting unit 101. For example, data 162 of a region wherein a
constant cross-section continues, which is an example of
continuity, is collected.
In this way, the relationship between the data 3 and the actual
world 1 is described with the Expression (5), and M pieces of data
162 are collected, thereby satisfying Expression (8), and the
actual world 1 can be estimated.
More specifically, in the event of N=M, the number of variables N
and the number of expressions M are equal, so the variables w.sub.i
can be obtained by formulating a simultaneous equation.
Also, in the event that N<M, various solving methods can be
applied. For example, the variables w.sub.i can be obtained by
least-square.
Now, the solving method by least-square will be described in
detail.
First, an Expression (9) for predicting data 3 from the actual
world 1 will be shown according to Expression (7).
'.function..times..times..function. ##EQU00006##
In Expression (9), P'.sub.j(x.sub.j, y.sub.j, t.sub.j) is a
prediction value.
The sum of squared differences E for the prediction value P' and
observed value P is represented by Expression (10).
.times..function.'.function. ##EQU00007##
The variables w.sub.i are obtained such that the sum of squared
differences E is the smallest. Accordingly, the partial
differential value of Expression (10) for each variable w.sub.k is
0. That is to say, Expression (11) holds.
.differential..differential..times..times..times..function..times..functi-
on..times..times..function. ##EQU00008##
Expression (11) yields Expression (12).
.times..function..times..times..times..function..times..function..times..-
function. ##EQU00009##
When Expression (12) holds with K=1 through N, the solution by
least-square is obtained. The normal equation thereof is shown in
Expression (13).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..times..times..times.-
.times..times..times..times..times..times..times..function..times..times..-
times..times..times..times..times..times..function..times..function..times-
..function..times..function..times..function..times..function..times..func-
tion..times..function..times..function..times..function..times..function..-
times..function..times..function..times..function..times..function..times.-
.function..times..function..times..function..times..function..times..funct-
ion..times..function..times..function..times..function..times..function..t-
imes..function. ##EQU00010##
From Expression (14) through Expression (16), Expression (13) can
be expressed as S.sub.MATW.sub.MAT=P.sub.MAT.
In Expression (13), S.sub.i represents the projection of the actual
world 1. In Expression (13), P.sub.j represents the data 3. In
Expression (13), w.sub.i represents variables for describing and
obtaining the characteristics of the signals of the actual world
1.
Accordingly, inputting the data 3 into Expression (13) and
obtaining W.sub.MAT by a matrix solution or the like enables the
actual world 1 to be estimated. That is to say, the actual world 1
can be estimated by computing Expression (17).
W.sub.MAT=S.sub.MAT.sup.-1P.sub.MAT (17)
Note that in the event that S.sub.MAT is not regular, a transposed
matrix of S.sub.MAT can be used to obtain W.sub.MAT.
The actual world estimating unit 102 estimates the actual world 1
by, for example, inputting the data 3 into Expression (13) and
obtaining W.sub.MAT by a matrix solution or the like.
Now, an even more detailed example will be described. For example,
the cross-sectional shape of the signals of the actual world 1,
i.e., the change in level as to the change in position, will be
described with a polynomial. Let us assume that the cross-sectional
shape of the signals of the actual world 1 is constant, and that
the cross-section of the signals of the actual world 1 moves at a
constant speed. Projection of the signals of the actual world 1
from the sensor 2 to the data 3 is formulated by three-dimensional
integration in the time-space direction of the signals of the
actual world 1.
The assumption that the cross-section of the signals of the actual
world 1 moves at a constant speed yields Expression (18) and
Expression (19).
dddd ##EQU00011##
Here, v.sub.x and v.sub.y are constant.
Using Expression (18) and Expression (19), the cross-sectional
shape of the signals of the actual world 1 can be represented as in
Expression (20). f(x',y')=f(x+v.sub.xt,y+v.sub.yt) (20)
Formulating projection of the signals of the actual world 1 from
the sensor 2 to the data 3 by three-dimensional integration in the
time-space direction of the signals of the actual world 1 yields
Expression (21).
.function..intg..times..intg..times..intg..times..function.''.times.d.tim-
es.d.times.d.intg..times..intg..times..intg..times..function..times..times-
..times.d.times.d.times.d ##EQU00012##
In Expression (21), S(x, y, t) represents an integrated value the
region from position x.sub.s to position x.sub.e for the spatial
direction X, from position y.sub.s to position y.sub.e for the
spatial direction Y, and from point-in-time t.sub.s to
point-in-time t.sub.e for the time direction t, i.e., the region
represented as a space-time cuboid.
Solving Expression (13) using a desired function f(x', y') whereby
Expression (21) can be determined enables the signals of the actual
world 1 to be estimated.
In the following, we will use the function indicated in Expression
(22) as an example of the function f(x', y').
.function.''.times.'.times.'.function..times..function..times.
##EQU00013##
That is to say, the signals of the actual world 1 are estimated to
include the continuity represented in Expression (18), Expression
(19), and Expression (22). This indicates that the cross-section
with a constant shape is moving in the space-time direction as
shown in FIG. 26.
Substituting Expression (22) into Expression (21) yields Expression
(23).
.function..times..intg..times..intg..times..intg..times..function..times.-
.times..times.d.times.d.times.d.times..times..times..times..function..time-
s..times..function..times..times..function..times..function..times..functi-
on. ##EQU00014##
wherein
Volume=(x.sub.e-x.sub.s) (y.sub.e-y.sub.s) (t.sub.e-t.sub.s)
S.sub.0 (x, y, t)=Volume/2.times.(x.sub.e+x.sub.s+v.sub.x
(t.sub.e+t.sub.s))
S.sub.1 (x, y, t)=Volume/2.times.(y.sub.e+y.sub.s+v.sub.y
(t.sub.e+t.sub.s))
S.sub.2(x, y, t)=1
holds.
FIG. 27 is a diagram illustrating an example of the M pieces of
data 162 extracted from the data 3. For example, let us say that 27
pixel values are extracted as the data 162, and that the extracted
pixel values are P.sub.j(x, y, t). In this case, j is 0 through
26.
In the example shown in FIG. 27, in the event that the pixel value
of the pixel corresponding to the position of interest at the
point-in-time t which is n is P.sub.13(x, y, t), and the direction
of array of the pixel values of the pixels having the continuity of
data (e.g., the direction in which the same-shaped pawl shapes
detected by the data continuity detecting unit 101 are arrayed) is
a direction connecting P.sub.4(X, y, t), P.sub.13(x, y, t), and
P.sub.22(x, y, t), the pixel values P.sub.9(x, y, t) through
P.sub.17(x, y, t) at the point-in-time t which is n, the pixel
values P.sub.0(x, y, t) through P.sub.8(x, y, t) at the
point-in-time t which is n-1 which is earlier in time than n, and
the pixel values P.sub.18(x, y, t) through P.sub.26(x, y, t) at the
point-in-time t which is n+1 which is later in time than n, are
extracted.
Now, the region regarding which the pixel values, which are the
data 3 output from the image sensor which is the sensor 2, have
been obtained, have a time-direction and two-dimensional spatial
direction expansion, as shown in FIG. 28. Now, as shown in FIG. 29,
the center of gravity of the cuboid corresponding to the pixel
values (the region regarding which the pixel values have been
obtained) can be used as the position of the pixel in the
space-time direction. The circle in FIG. 29 indicates the center of
gravity.
Generating Expression (13) from the 27 pixel values P.sub.0(x, y,
t) through P.sub.26(x, y, t) and from Expression (23), and
obtaining W, enables the actual world 1 to be estimated.
In this way, the actual world estimating unit 102 generates
Expression (13) from the 27 pixel values P.sub.0(x, y, t) through
P.sub.26(x, y, t) and from Expression (23), and obtains W, thereby
estimating the signals of the actual world 1.
Note that a Gaussian function, a sigmoid function, or the like, can
be used for the function f.sub.i(x, y, t).
An example of processing for generating high-resolution data 181
with even higher resolution, corresponding to the data 3, from the
estimated actual world 1 signals, will be described with reference
to FIG. 30 through FIG. 34.
As shown in FIG. 30, the data 3 has a value wherein signals of the
actual world 1 are integrated in the time direction and
two-dimensional spatial directions. For example, a pixel value
which is data 3 that has been output from the image sensor which is
the sensor 2 has a value wherein the signals of the actual world 1,
which is light cast into the detecting device, are integrated by
the shutter time which is the detection time in the time direction,
and integrated by the photoreception region of the detecting
element in the spatial direction.
Conversely, as shown in FIG. 31, the high-resolution data 181 with
even higher resolution in the spatial direction is generated by
integrating the estimated actual world 1 signals in the time
direction by the same time as the detection time of the sensor 2
which has output the data 3, and also integrating in the spatial
direction by a region narrower in comparison with the
photoreception region of the detecting element of the sensor 2
which has output the data 3.
Note that at the time of generating the high-resolution data 181
with even higher resolution in the spatial direction, the region
where the estimated signals of the actual world 1 are integrated
can be set completely disengaged from photoreception region of the
detecting element of the sensor 2 which has output the data 3. For
example, the high-resolution data 181 can be provided with
resolution which is that of the data 3 magnified in the spatial
direction by an integer, of course, and further, can be provided
with resolution which is that of the data 3 magnified in the
spatial direction by a rational number such as 5/3 times, for
example.
Also, as shown in FIG. 32, the high-resolution data 181 with even
higher resolution in the time direction is generated by integrating
the estimated actual world 1 signals in the spatial direction by
the same region as the photoreception region of the detecting
element of the sensor 2 which has output the data 3, and also
integrating in the time direction by a time shorter than the
detection time of the sensor 2 which has output the data 3.
Note that at the time of generating the high-resolution data 181
with even higher resolution in the time direction, the time by
which the estimated signals of the actual world 1 are integrated
can be set completely disengaged from shutter time of the detecting
element of the sensor 2 which has output the data 3. For example,
the high-resolution data 181 can be provided with resolution which
is that of the data 3 magnified in the time direction by an
integer, of course, and further, can be provided with resolution
which is that of the data 3 magnified in the time direction by a
rational number such as 7/4 times, for example.
As shown in FIG. 33, high-resolution data 181 with movement
blurring removed is generated by integrating the estimated actual
world 1 signals only in the spatial direction and not in the time
direction.
Further, as shown in FIG. 34, high-resolution data 181 with higher
resolution in the time direction and space direction is generated
by integrating the estimated actual world 1 signals in the spatial
direction by a region narrower in comparison with the
photoreception region of the detecting element of the sensor 2
which has output the data 3, and also integrating in the time
direction by a time shorter in comparison with the detection time
of the sensor 2 which has output the data 3.
In this case, the region and time for integrating the estimated
actual world 1 signals can be set completely unrelated to the
photoreception region and shutter time of the detecting element of
the sensor 2 which has output the data 3.
Thus, the image generating unit 103 generates data with higher
resolution in the time direction or the spatial direction, by
integrating the estimated actual world 1 signals by a desired
space-time region, for example.
Accordingly, data which is more accurate with regard to the signals
of the actual world 1, and which has higher resolution in the time
direction or the space direction, can be generated by estimating
the signals of the actual world 1.
An example of an input image and the results of processing with the
signal processing device 4 according to the present invention will
be described with reference to FIG. 35 through FIG. 39.
FIG. 35 is a diagram illustrating an original image of an input
image. FIG. 36 is a diagram illustrating an example of an input
image. The input image shown in FIG. 36 is an image generated by
taking the average value of pixel values of pixels belonging to
blocks made up of 2 by 2 pixels of the image shown in FIG. 35, as
the pixel value of a single pixel. That is to say, the input image
is an image obtained by applying spatial direction integration to
the image shown in FIG. 35, imitating the integrating properties of
the sensor.
The original image shown in FIG. 35 contains an image of a fine
line inclined at approximately 5 degrees in the clockwise direction
from the vertical direction. In the same way, the input image shown
in FIG. 36 contains an image of a fine line inclined at
approximately 5 degrees in the clockwise direction from the
vertical direction.
FIG. 37 is a diagram illustrating an image obtained by applying
conventional class classification adaptation processing to the
input image shown in FIG. 36. Now, class classification processing
is made up of class classification processing and adaptation
processing, wherein the data is classified based on the nature
thereof by the class classification adaptation processing, and
subjected to adaptation processing for each class. In the
adaptation processing, a low-image quality or standard image
quality image, for example, is converted into a high image quality
image by being subjected to mapping (mapping) using a predetermined
tap coefficient.
It can be understood in the image shown in FIG. 37 that the image
of the fine line is different to that of the original image in FIG.
35.
FIG. 38 is a diagram illustrating the results of detecting the fine
line regions from the input image shown in the example in FIG. 36,
by the data continuity detecting unit 101. In FIG. 38, the white
region indicates the fine line region, i.e., the region wherein the
arc shapes shown in FIG. 14 are arrayed.
FIG. 39 is a diagram illustrating an example of the output image
output from the signal processing device 4 according to the present
invention, with the image shown in FIG. 36 as the input image. As
shown in FIG. 39, the signals processing device 4 according to the
present invention yields an image closer to the fine line image of
the original image shown in FIG. 35.
FIG. 40 is a flowchart for describing the processing of signals
with the signal processing device 4 according to the present
invention.
In step S101, the data continuity detecting unit 101 executes the
processing for detecting continuity. The data continuity detecting
unit 101 detects data continuity contained in the input image which
is the data 3, and supplies the data continuity information
indicating the detected data continuity to the actual world
estimating unit 102 and the image generating unit 103.
The data continuity detecting unit 101 detects the continuity of
data corresponding to the continuity of the signals of the actual
world. In the processing in step S101, the continuity of data
detected by the data continuity detecting unit 101 is either part
of the continuity of the image of the actual world 1 contained in
the data 3, or continuity which has changed from the continuity of
the signals of the actual world 1.
The data continuity detecting unit 101 detects the data continuity
by detecting a region having a constant characteristic in a
predetermined dimensional direction. Also, the data continuity
detecting unit 101 detects data continuity by detecting angle
(gradient) in the spatial direction indicating the an array of the
same shape.
Details of the continuity detecting processing in step S101 will be
described later.
Note that the data continuity information can be used as features,
indicating the characteristics of the data 3.
In step S102, the actual world estimating unit 102 executes
processing for estimating the actual world. That is to say, the
actual world estimating unit 102 estimates the signals of the
actual world based on the input image and the data continuity
information supplied from the data continuity detecting unit 101.
In the processing in step S102 for example, the actual world
estimating unit 102 estimates the signals of the actual world 1 by
predicting a model 161 approximating (describing) the actual world
1. The actual world estimating unit 102 supplies the actual world
estimation information indicating the estimated signals of the
actual world 1 to the image generating unit 103.
For example, the actual world estimating unit 102 estimates the
actual world 1 signals by predicting the width of the linear
object. Also, for example, the actual world estimating unit 102
estimates the actual world 1 signals by predicting a level
indicating the color of the linear object.
Details of processing for estimating the actual world in step S102
will be described later.
Note that the actual world estimation information can be used as
features, indicating the characteristics of the data 3.
In step S103, the image generating unit 103 performs image
generating processing, and the processing ends. That is to say, the
image generating unit 103 generates an image based on the actual
world estimation information, and outputs the generated image. Or,
the image generating unit 103 generates an image based on the data
continuity information and actual world estimation information, and
outputs the generated image.
For example, in the processing in step S103, the image generating
unit 103 integrates a function approximating the generated real
world light signals in the spatial direction, based on the actual
world estimated information, hereby generating an image with higher
resolution in the spatial direction in comparison with the input
image, and outputs the generated image. For example, the image
generating unit 103 integrates a function approximating the
generated real world light signals in the time-space direction,
based on the actual world estimated information, hereby generating
an image with higher resolution in the time direction and the
spatial direction in comparison with the input image, and outputs
the generated image. The details of the image generating processing
in step S103 will be described later.
Thus, the signal processing device 4 according to the present
invention detects data continuity from the data 3, and estimates
the actual world 1 from the detected data continuity. The signal
processing device 4 then generates signals closer approximating the
actual world 1 based on the estimated actual world 1.
As described above, in the event of performing the processing for
estimating signals of the real world, accurate and highly-precise
processing results can be obtained.
Also, in the event that first signals which are real world signals
having first dimensions are projected, the continuity of data
corresponding to the lost continuity of the real world signals is
detected for second signals of second dimensions, having a number
of dimensions fewer than the first dimensions, from which a part of
the continuity of the signals of the real world has been lost, and
the first signals are estimated by estimating the lost real world
signals continuity based on the detected data continuity, accurate
and highly-precise processing results can be obtained as to the
events in the real world.
Next, the details of the configuration of the data continuity
detecting unit 101 will be described.
FIG. 41 is a block diagram illustrating the configuration of the
data continuity detecting unit 101.
Upon taking an image of an object which is a fine line, the data
continuity detecting unit 101, of which the configuration is shown
in FIG. 41, detects the continuity of data contained in the data 3,
which is generated from the continuity in that the cross-sectional
shape which the object has is the same. That is to say, the data
continuity detecting unit 101 of the configuration shown in FIG. 41
detects the continuity of data contained in the data 3, which is
generated from the continuity in that the change in level of light
as to the change in position in the direction orthogonal to the
length-wise direction is the same at an arbitrary position in the
length-wise direction, which the image of the actual world 1 which
is a fine line, has.
More specifically, the data continuity detecting unit 101 of which
configuration is shown in FIG. 41 detects the region where multiple
arc shapes (half-disks) having a predetermined length are arrayed
in a diagonally-offset adjacent manner, within the data 3 obtained
by taking an image of a fine line with the sensor 2 having spatial
integration effects.
The data continuity detecting unit 101 extracts the portions of the
image data other than the portion of the image data where the image
of the fine line having data continuity has been projected
(hereafter, the portion of the image data where the image of the
fine line having data continuity has been projected will also be
called continuity component, and the other portions will be called
non-continuity component), from an input image which is the data 3,
detects the pixels where the image of the fine line of the actual
world 1 has been projected, from the extracted non-continuity
component and the input image, and detects the region of the input
image made up of pixels where the image of the fine line of the
actual world 1 has been projected.
A non-continuity component extracting unit 201 extracts the
non-continuity component from the input image, and supplies the
non-continuity component information indicating the extracted
non-continuity component to a peak detecting unit 202 and a
monotonous increase/decrease detecting unit 203 along with the
input image.
For example, as shown in FIG. 42, in the event that an image of the
actual world 1 wherein a fine line exists in front of a background
with an approximately constant light level is projected on the data
3, the non-continuity component extracting unit 201 extracts the
non-continuity component which is the background, by approximating
the background in the input image which is the data 3, on a plane,
as shown in FIG. 43. In FIG. 43, the solid line indicates the pixel
values of the data 3, and the dotted line illustrates the
approximation values indicated by the plane approximating the
background. In FIG. 43, A denotes the pixel value of the pixel
where the image of the fine line has been projected, and the PL
denotes the plane approximating the background.
In this way, the pixel values of the multiple pixels at the portion
of the image data having data continuity are discontinuous as to
the non-continuity component.
The non-continuity component extracting unit 201 detects the
discontinuous portion of the pixel values of the multiple pixels of
the image data which is the data 3, where an image which is light
signals of the actual world 1 has been projected and a part of the
continuity of the image of the actual world 1 has been lost.
Details of the processing for extracting the non-continuity
component with the non-continuity component extracting unit 201
will be described later.
The peak detecting unit 202 and the monotonous increase/decrease
detecting unit 203 remove the non-continuity component from the
input image, based on the non-continuity component information
supplied from the non-continuity component extracting unit 201. For
example, the peak detecting unit 202 and the monotonous
increase/decrease detecting unit 203 remove the non-continuity
component from the input image by setting the pixel values of the
pixels of the input image where only the background image has been
projected, to 0. Also, for example, the peak detecting unit 202 and
the monotonous increase/decrease detecting unit 203 remove the
non-continuity component from the input image by subtracting values
approximated by the plane PL from the pixel values of each pixel of
the input image.
Since the background can be removed from the input image, the peak
detecting unit 202 through continuousness detecting unit 204 can
process only the portion of the image data where the fine line has
be projected, thereby further simplifying the processing by the
peak detecting unit 202 through the continuousness detecting unit
204.
Note that the non-continuity component extracting unit 201 may
supply image data wherein the non-continuity component has been
removed form the input image, to the peak detecting unit 202 and
the monotonous increase/decrease detecting unit 203.
In the example of processing described below, the image data
wherein the non-continuity component has been removed from the
input image, i.e., image data made up from only pixel containing
the continuity component, is the object.
Now, description will be made regarding the image data upon which
the fine line image has been projected, which the peak detecting
unit 202 through continuousness detecting unit 204 are to
detect.
In the event that there is no optical LPF, the cross-dimensional
shape in the spatial direction Y (change in the pixel values as to
change in the position in the spatial direction) of the image data
upon which the fine line image has been projected as shown in FIG.
42 can be thought to be the trapezoid shown in FIG. 44, or the
triangle shown in FIG. 45. However, ordinary image sensors have an
optical LPF with the image sensor obtaining the image which has
passed through the optical LPF and projects the obtained image on
the data 3, so in reality, the cross-dimensional shape of the image
data with fine lines in the spatial direction Y has a shape
resembling Gaussian distribution, as shown in FIG. 46.
The peak detecting unit 202 through continuousness detecting unit
204 detect a region made up of pixels upon which the fine line
image has been projected wherein the same cross-sectional shape
(change in the pixel values as to change in the position in the
spatial direction) is arrayed vertically in the screen at constant
intervals, and further, detect a region made up of pixels upon
which the fine line image has been projected which is a region
having data continuity, by detecting regional connection
corresponding to the length-wise direction of the fine line of the
actual world 1. That is to say, the peak detecting unit 202 through
continuousness detecting unit 204 detect regions wherein arc shapes
(half-disc shapes) are formed on a single vertical row of pixels in
the input image, and determine whether or not the detected regions
are adjacent in the horizontal direction, thereby detecting
connection of regions where arc shapes are formed, corresponding to
the length-wise direction of the fine line image which is signals
of the actual world 1.
Also, the peak detecting unit 202 through continuousness detecting
unit 204 detect a region made up of pixels upon which the fine line
image has been projected wherein the same cross-sectional shape is
arrayed horizontally in the screen at constant intervals, and
further, detect a region made up of pixels upon which the fine line
image has been projected which is a region having data continuity,
by detecting connection of detected regions corresponding to the
length-wise direction of the fine line of the actual world 1. That
is to say, the peak detecting unit 202 through continuousness
detecting unit 204 detect regions wherein arc shapes are formed on
a single horizontal row of pixels in the input image, and determine
whether or not the detected regions are adjacent in the vertical
direction, thereby detecting connection of regions where arc shapes
are formed, corresponding to the length-wise direction of the fine
line image, which is signals of the actual world 1.
First, description will be made regarding processing for detecting
a region of pixels upon which the fine line image has been
projected wherein the same arc shape is arrayed vertically in the
screen at constant intervals.
The peak detecting unit 202 detects a pixel having a pixel value
greater than the surrounding pixels, i.e., a peak, and supplies
peak information indicating the position of the peak to the
monotonous increase/decrease detecting unit 203. In the event that
pixels arrayed in a single vertical row in the screen are the
object, the peak detecting unit 202 compares the pixel value of the
pixel position upwards in the screen and the pixel value of the
pixel position downwards in the screen, and detects the pixel with
the greater pixel value as the peak. The peak detecting unit 202
detects one or multiple peaks from a single image, e.g., from the
image of a single frame.
A single screen contains frames or fields. This holds true in the
following description as well.
For example, the peak detecting unit 202 selects a pixel of
interest from pixels of an image of one frame which have not yet
been taken as pixels of interest, compares the pixel value of the
pixel of interest with the pixel value of the pixel above the pixel
of interest, compares the pixel value of the pixel of interest with
the pixel value of the pixel below the pixel of interest, detects a
pixel of interest which has a greater pixel value than the pixel
value of the pixel above and a greater pixel value than the pixel
value of the pixel below, and takes the detected pixel of interest
as a peak. The peak detecting unit supplies peak information
indicating the detected peak to the monotonous increase/decrease
detecting unit 203.
There are cases wherein the peak detecting unit 202 does not detect
a peak. For example, in the event that the pixel values of all of
the pixels of an image are the same value, or in the event that the
pixel values decrease in one or two directions, no peak is
detected. In this case, no fine line image has been projected on
the image data.
The monotonous increase/decrease detecting unit 203 detects a
candidate for a region made up of pixels upon which the fine line
image has been projected wherein the pixels are vertically arrayed
in a single row as to the peak detected by the peak detecting unit
202, based upon the peak information indicating the position of the
peak supplied from the peak detecting unit 202, and supplies the
region information indicating the detected region to the
continuousness detecting unit 204 along with the peak
information.
More specifically, the monotonous increase/decrease detecting unit
203 detects a region made up of pixels having pixel values
monotonously decreasing with reference to the peak pixel value, as
a candidate of a region made up of pixels upon which the image of
the fine line has been projected. Monotonous decrease means that
the pixel values of pixels which are farther distance-wise from the
peak are smaller than the pixel values of pixels which are closer
to the peak.
Also, the monotonous increase/decrease detecting unit 203 detects a
region made up of pixels having pixel values monotonously
increasing with reference to the peak pixel value, as a candidate
of a region made up of pixels upon which the image of the fine line
has been projected. Monotonous increase means that the pixel values
of pixels which are farther distance-wise from the peak are greater
than the pixel values of pixels which are closer to the peak.
In the following, the processing regarding regions of pixels having
pixel values monotonously increasing is the same as the processing
regarding regions of pixels having pixel values monotonously
decreasing, so description thereof will be omitted. Also, with the
description regarding processing for detecting a region of pixels
upon which the fine line image has been projected wherein the same
arc shape is arrayed horizontally in the screen at constant
intervals, the processing regarding regions of pixels having pixel
values monotonously increasing is the same as the processing
regarding regions of pixels having pixel values monotonously
decreasing, so description thereof will be omitted.
For example, the monotonous increase/decrease detecting unit 203
detects pixel values of each of the pixels in a vertical row as to
a peak, the difference as to the pixel value of the pixel above,
and the difference as to the pixel value of the pixel below. The
monotonous increase/decrease detecting unit 203 then detects a
region wherein the pixel value monotonously decreases by detecting
pixels wherein the sign of the difference changes.
Further, the monotonous increase/decrease detecting unit 203
detects, from the region wherein pixel values monotonously
decrease, a region made up of pixels having pixel values with the
same sign as that of the pixel value of the peak, with the sign of
the pixel value of the peak as a reference, as a candidate of a
region made up of pixels upon which the image of the fine line has
been projected.
For example, the monotonous increase/decrease detecting unit 203
compares the sign of the pixel value of each pixel with the sign of
the pixel value of the pixel above and sign of the pixel value of
the pixel below, and detects the pixel where the sign of the pixel
value changes, thereby detecting a region of pixels having pixel
values of the same sign as the peak within the region where pixel
values monotonously decrease.
Thus, the monotonous increase/decrease detecting unit 203 detects a
region formed of pixels arrayed in a vertical direction wherein the
pixel values monotonously decrease as to the peak and have pixels
values of the same sign as the peak.
FIG. 47 is a diagram describing processing for peak detection and
monotonous increase/decrease region detection, for detecting the
region of pixels wherein the image of the fine line has been
projected, from the pixel values as to a position in the spatial
direction Y.
In FIG. 47 through FIG. 49, P represents a peak. In the description
of the data continuity detecting unit 101 of which the
configuration is shown in FIG. 41, P represents a peak.
The peak detecting unit 202 compares the pixel values of the pixels
with the pixel values of the pixels adjacent thereto in the spatial
direction Y, and detects the peak P by detecting a pixel having a
pixel value greater than the pixel values of the two pixels
adjacent in the spatial direction Y.
The region made up of the peak P and the pixels on both sides of
the peak P in the spatial direction Y is a monotonous decrease
region wherein the pixel values of the pixels on both sides in the
spatial direction Y monotonously decrease as to the pixel value of
the peak P. In FIG. 47, the arrow denoted A and the arrow denoted
by B represent the monotonous decrease regions existing on either
side of the peak P.
The monotonous increase/decrease detecting unit 203 obtains the
difference between the pixel values of each pixel and the pixel
values of the pixels adjacent in the spatial direction Y, and
detects pixels where the sign of the difference changes. The
monotonous increase/decrease detecting unit 203 takes the boundary
between the detected pixel where the sign of the difference changes
and the pixel immediately prior thereto (on the peak P side) as the
boundary of the fine line region made up of pixels where the image
of the fine line has been projected.
In FIG. 47, the boundary of the fine line region which is the
boundary between the pixel where the sign of the difference changes
and the pixel immediately prior thereto (on the peak P side) is
denoted by C.
Further, the monotonous increase/decrease detecting unit 203
compares the sign of the pixel values of each pixel with the pixel
values of the pixels adjacent thereto in the spatial direction Y,
and detects pixels where the sign of the pixel value changes in the
monotonous decrease region. The monotonous increase/decrease
detecting unit 203 takes the boundary between the detected pixel
where the sign of the pixel value changes and the pixel immediately
prior thereto (on the peak P side) as the boundary of the fine line
region.
In FIG. 47, the boundary of the fine line region which is the
boundary between the pixel where the sign of the pixel value
changes and the pixel immediately prior thereto (on the peak P
side) is denoted by P.
As shown in FIG. 47, the fine line region F made up of pixels where
the image of the fine line has been projected is the region between
the fine line region boundary C and the fine line region boundary
D.
The monotonous increase/decrease detecting unit 203 obtains a fine
line region F which is longer than a predetermined threshold, from
fine line regions F made up of such monotonous increase/decrease
regions, i.e., a fine line region F having a greater number of
pixels than the threshold value. For example, in the event that the
threshold value is 3, the monotonous increase/decrease detecting
unit 203 detects a fine line region F including 4 or more
pixels.
Further, the monotonous increase/decrease detecting unit 203
compares the pixel value of the peak P, the pixel value of the
pixel to the right side of the peak P, and the pixel value of the
pixel to the left side of the peak P, from the fine line region F
thus detected, each with the threshold value, detects a fine pixel
region F having the peak P wherein the pixel value of the peak P
exceeds the threshold value, and wherein the pixel value of the
pixel to the right side of the peak P is the threshold value or
lower, and wherein the pixel value of the pixel to the left side of
the peak P is the threshold value or lower, and takes the detected
fine line region F as a candidate for the region made up of pixels
containing the component of the fine line image.
In other words, determination is made that a fine line region F
having the peak P, wherein the pixel value of the peak P is the
threshold value or lower, or wherein the pixel value of the pixel
to the right side of the peak P exceeds the threshold value, or
wherein the pixel value of the pixel to the left side of the peak P
exceeds the threshold value, does not contain the component of the
fine line image, and is eliminated from candidates for the region
made up of pixels including the component of the fine line
image.
That is, as shown in FIG. 48, the monotonous increase/decrease
detecting unit 203 compares the pixel value of the peak P with the
threshold value, and also compares the pixel value of the pixel
adjacent to the peak P in the spatial direction X (the direction
indicated by the dotted line AA') with the threshold value, thereby
detecting the fine line region F to which the peak P belongs,
wherein the pixel value of the peak P exceeds the threshold value
and wherein the pixel values of the pixel adjacent thereto in the
spatial direction X are equal to or below the threshold value.
FIG. 49 is a diagram illustrating the pixel values of pixels
arrayed in the spatial direction X indicated by the dotted line AA'
in FIG. 48. The fine line region F to which the peak P belongs,
wherein the pixel value of the peak P exceeds the threshold value
Th.sub.s and wherein the pixel values of the pixel adjacent thereto
in the spatial direction X are equal to or below the threshold
value Th.sub.s, contains the fine line component.
Note that an arrangement may be made wherein the monotonous
increase/decrease detecting unit 203 compares the difference
between the pixel value of the peak P and the pixel value of the
background with the threshold value, taking the pixel value of the
background as a reference, and also compares the difference between
the pixel value of the pixels adjacent to the peak P in the spatial
direction and the pixel value of the background with the threshold
value, thereby detecting the fine line region F to which the peak P
belongs, wherein the difference between the pixel value of the peak
P and the pixel value of the background exceeds the threshold
value, and wherein the difference between the pixel value of the
pixel adjacent in the spatial direction X and the pixel value of
the background is equal to or below the threshold value.
The monotonous increase/decrease detecting unit 203 outputs to the
continuousness detecting unit 204 monotonous increase/decrease
region information indicating a region made up of pixels of which
the pixel value monotonously decrease with the peak P as a
reference and the sign of the pixel value is the same as that of
the peak P, wherein the peak P exceeds the threshold value and
wherein the pixel value of the pixel to the right side of the peak
P is equal to or below the threshold value and the pixel value of
the pixel to the left side of the peak P is equal to or below the
threshold value.
In the event of detecting a region of pixels arrayed in a single
row in the vertical direction of the screen where the image of the
fine line has been projected, pixels belonging to the region
indicated by the monotonous increase/decrease region information
are arrayed in the vertical direction and include pixels where the
image of the fine line has been projected. That is to say, the
region indicated by the monotonous increase/decrease region
information includes a region formed of pixels arrayed in a single
row in the vertical direction of the screen where the image of the
fine line has been projected.
In this way, the apex detecting unit 202 and the monotonous
increase/decrease detecting unit 203 detects a continuity region
made up of pixels where the image of the fine line has been
projected, employing the nature that, of the pixels where the image
of the fine line has been projected, change in the pixel values in
the spatial direction Y approximates Gaussian distribution.
Of the region made up of pixels arrayed in the vertical direction,
indicated by the monotonous increase/decrease region information
supplied from the monotonous increase/decrease detecting unit 203,
the continuousness detecting unit 204 detects regions including
pixels adjacent in the horizontal direction, i.e., regions having
similar pixel value change and duplicated in the vertical
direction, as continuous regions, and outputs the peak information
and data continuity information indicating the detected continuous
regions. The data continuity information includes monotonous
increase/decrease region information, information indicating the
connection of regions, and so forth.
Arc shapes are aligned at constant intervals in an adjacent manner
with the pixels where the fine line has been projected, so the
detected continuous regions include the pixels where the fine line
has been projected.
The detected continuous regions include the pixels where arc shapes
are aligned at constant intervals in an adjacent manner to which
the fine line has been projected, so the detected continuous
regions are taken as a continuity region, and the continuousness
detecting unit 204 outputs data continuity information indicating
the detected continuous regions.
That is to say, the continuousness detecting unit 204 uses the
continuity wherein arc shapes are aligned at constant intervals in
an adjacent manner in the data 3 obtained by imaging the fine line,
which has been generated due to the continuity of the image of the
fine line in the actual world 1, the nature of the continuity being
continuing in the length direction, so as to further narrow down
the candidates of regions detected with the peak detecting unit 202
and the monotonous increase/decrease detecting unit 203.
FIG. 50 is a diagram describing the processing for detecting the
continuousness of monotonous increase/decrease regions.
As shown in FIG. 50, in the event that a fine line region F formed
of pixels aligned in a single row in the vertical direction of the
screen includes pixels adjacent in the horizontal direction, the
continuousness detecting unit 204 determines that there is
continuousness between the two monotonous increase/decrease
regions, and in the event that pixels adjacent in the horizontal
direction are not included, determines that there is no
continuousness between the two fine line regions F. For example, a
fine line region F.sub.-1 made up of pixels aligned in a single row
in the vertical direction of the screen is determined to be
continuous to a fine line region F.sub.0 made up of pixels aligned
in a single row in the vertical direction of the screen in the
event of containing a pixel adjacent to a pixel of the fine line
region F.sub.0 in the horizontal direction. The fine line region
F.sub.0 made up of pixels aligned in a single row in the vertical
direction of the screen is determined to be continuous to a fine
line region F.sub.l made up of pixels aligned in a single row in
the vertical direction of the screen in the event of containing a
pixel adjacent to a pixel of the fine line region F.sub.l in the
horizontal direction.
In this way, regions made up of pixels aligned in a single row in
the vertical direction of the screen where the image of the fine
line has been projected are detected by the peak detecting unit 202
through the continuousness detecting unit 204.
As described above, the peak detecting unit 202 through the
continuousness detecting unit 204 detect regions made up of pixels
aligned in a single row in the vertical direction of the screen
where the image of the fine line has been projected, and further
detect regions made up of pixels aligned in a single row in the
horizontal direction of the screen where the image of the fine line
has been projected.
Note that the order of processing does not restrict the present
invention, and may be executed in parallel, as a matter of
course.
That is to say, the peak detecting unit 202, with regard to of
pixels aligned in a single row in the horizontal direction of the
screen, detects as a peak a pixel which has a pixel value greater
in comparison with the pixel value of the pixel situated to the
left side on the screen and the pixel value of the pixel situated
to the right side on the screen, and supplies peak information
indicating the position of the detected peak to the monotonous
increase/decrease detecting unit 203. The peak detecting unit 202
detects one or multiple peaks from one image, for example, one
frame image.
For example, the peak detecting unit 202 selects a pixel of
interest from pixels in the one frame image which has not yet been
taken as a pixel of interest, compares the pixel value of the pixel
of interest with the pixel value of the pixel to the left side of
the pixel of interest, compares the pixel value of the pixel of
interest with the pixel value of the pixel to the right side of the
pixel of interest, detects a pixel of interest having a pixel value
greater than the pixel value of the pixel to the left side of the
pixel of interest and having a pixel value greater than the pixel
value of the pixel to the right side of the pixel of interest, and
takes the detected pixel of interest as a peak. The peak detecting
unit 202 supplies peak information indicating the detected peak to
the monotonous increase/decrease detecting unit 203.
There are cases wherein the peak detecting unit 202 does not detect
a peak.
The monotonous increase/decrease detecting unit 203 detects
candidates for a region made up of pixels aligned in a single row
in the horizontal direction as to the peak detected by the peak
detecting unit 202 wherein the fine line image has been projected,
and supplies the monotonous increase/decrease region information
indicating the detected region to the continuousness detecting unit
204 along with the peak information.
More specifically, the monotonous increase/decrease detecting unit
203 detects regions made up of pixels having pixel values
monotonously decreasing with the pixel value of the peak as a
reference, as candidates of regions made up of pixels where the
fine line image has been projected.
For example, the monotonous increase/decrease detecting unit 203
obtains, with regard to each pixel in a single row in the
horizontal direction as to the peak, the pixel value of each pixel,
the difference as to the pixel value of the pixel to the left side,
and the difference as to the pixel value of the pixel to the right
side. The monotonous increase/decrease detecting unit 203 then
detects the region where the pixel value monotonously decreases by
detecting the pixel where the sign of the difference changes.
Further, the monotonous increase/decrease detecting unit 203
detects a region made up of pixels having pixel values with the
same sign as the pixel value as the sign of the pixel value of the
peak, with reference to the sign of the pixel value of the peak, as
a candidate for a region made up of pixels where the fine line
image has been projected.
For example, the monotonous increase/decrease detecting unit 203
compares the sign of the pixel value of each pixel with the sign of
the pixel value of the pixel to the left side or with the sign of
the pixel value of the pixel to the right side, and detects the
pixel where the sign of the pixel value changes, thereby detecting
a region made up of pixels having pixel values with the same sign
as the peak, from the region where the pixel values monotonously
decrease.
Thus, the monotonous increase/decrease detecting unit 203 detects a
region made up of pixels aligned in the horizontal direction and
having pixel values with the same sign as the peak wherein the
pixel values monotonously decrease as to the peak.
From a fine line region made up of such a monotonous
increase/decrease region, the monotonous increase/decrease
detecting unit 203 obtains a fine line region longer than a
threshold value set beforehand, i.e., a fine line region having a
greater number of pixels than the threshold value.
Further, from the fine line region thus detected, the monotonous
increase/decrease detecting unit 203 compares the pixel value of
the peak, the pixel value of the pixel above the peak, and the
pixel value of the pixel below the peak, each with the threshold
value, detects a fine line region to which belongs a peak wherein
the pixel value of the peak exceeds the threshold value, the pixel
value of the pixel above the peak is within the threshold, and the
pixel value of the pixel below the peak is within the threshold,
and takes the detected fine line region as a candidate for a region
made up of pixels containing the fine line image component.
Another way of saying this is that fine line regions to which
belongs a peak wherein the pixel value of the peak is within the
threshold value, or the pixel value of the pixel above the peak
exceeds the threshold, or the pixel value of the pixel below the
peak exceeds the threshold, are determined to not contain the fine
line image component, and are eliminated from candidates of the
region made up of pixels containing the fine line image
component.
Note that the monotonous increase/decrease detecting unit 203 may
be arranged to take the background pixel value as a reference,
compare the difference between the pixel value of the pixel and the
pixel value of the background with the threshold value, and also to
compare the difference between the pixel value of the background
and the pixel values adjacent to the peak in the vertical direction
with the threshold value, and take a detected fine line region
wherein the difference between the pixel value of the peak and the
pixel value of the background exceeds the threshold value, and the
difference between the pixel value of the background and the pixel
value of the pixels adjacent in the vertical direction is within
the threshold, as a candidate for a region made up of pixels
containing the fine line image component.
The monotonous increase/decrease detecting unit 203 supplies to the
continuousness detecting unit 204 monotonous increase/decrease
region information indicating a region made up of pixels having a
pixel value sign which is the same as the peak and monotonously
decreasing pixel values as to the peak as a reference, wherein the
peak exceeds the threshold value, and the pixel value of the pixel
to the right side of the peak is within the threshold, and the
pixel value of the pixel to the left side of the peak is within the
threshold.
In the event of detecting a region made up of pixels aligned in a
single row in the horizontal direction of the screen wherein the
image of the fine line has been projected, pixels belonging to the
region indicated by the monotonous increase/decrease region
information include pixels aligned in the horizontal direction
wherein the image of the fine line has been projected. That is to
say, the region indicated by the monotonous increase/decrease
region information includes a region made up of pixels aligned in a
single row in the horizontal direction of the screen wherein the
image of the fine line has been projected.
Of the regions made up of pixels aligned in the horizontal
direction indicated in the monotonous increase/decrease region
information supplied from the monotonous increase/decrease
detecting unit 203, the continuousness detecting unit 204 detects
regions including pixels adjacent in the vertical direction, i.e.,
regions having similar pixel value change and which are repeated in
the horizontal direction, as continuous regions, and outputs data
continuity information indicating the peak information and the
detected continuous regions. The data continuity information
includes information indicating the connection of the regions.
At the pixels where the fine line has been projected, arc shapes
are arrayed at constant intervals in an adjacent manner, so the
detected continuous regions include pixels where the fine line has
been projected.
The detected continuous regions include pixels where arc shapes are
arrayed at constant intervals wherein the fine line has been
projected, so the detected continuous regions are taken as a
continuity region, and the continuousness detecting unit 204
outputs data continuity information indicating the detected
continuous regions.
That is to say, the continuousness detecting unit 204 uses the
continuity which is that the arc shapes are arrayed at constant
intervals in an adjacent manner in the data 3 obtained by imaging
the fine line, generated from the continuity of the image of the
fine line in the actual world 1 which is continuation in the length
direction, so as to further narrow down the candidates of regions
detected by the peak detecting unit 202 and the monotonous
increase/decrease detecting unit 203.
FIG. 51 is a diagram illustrating an example of an image wherein
the continuity component has been extracted by planar
approximation. FIG. 52 is a diagram illustrating the results of
detecting peaks in the image shown in FIG. 51, and detecting
monotonously decreasing regions. In FIG. 52, the portions indicated
by white are the detected regions.
FIG. 53 is a diagram illustrating regions wherein continuousness
has been detected by detecting continuousness of adjacent regions
in the image shown in FIG. 52. In FIG. 53, the portions shown in
white are regions where continuity has been detected. It can be
understood that detection of continuousness further identifies the
regions.
FIG. 54 is a diagram illustrating the pixel values of the regions
shown in FIG. 53, i.e., the pixel values of the regions where
continuousness has been detected.
Thus, the data continuity detecting unit 101 is capable of
detecting continuity contained in the data 3 which is the input
image. That is to say, the data continuity detecting unit 101 can
detect continuity of data included in the data 3 which has been
generated by the actual world 1 image which is a fine line having
been projected on the data 3. The data continuity detecting unit
101 detects, from the data 3, regions made up of pixels where the
actual world 1 image which is a fine line has been projected.
FIG. 55 is a diagram illustrating an example of other processing
for detecting regions having continuity, where a fine line image
has been projected, with the data continuity detecting unit
101.
As shown in FIG. 55, the data continuity detecting unit 101
calculates the absolute value of difference of pixel values for
each pixel and adjacent pixels. The calculated absolute values of
difference are placed corresponding to the pixels. For example, in
a situation such as shown in FIG. 55 wherein there are pixels
aligned which have respective pixel values of P0, P1, and P2, the
data continuity detecting unit 101 calculates the difference
d0=P0-P1 and the difference d1=P1-P2. Further, the data continuity
detecting unit 101 calculates the absolute values of the difference
d0 and the difference d1.
In the event that the non-continuity component contained in the
pixel values P0, P1, and P2 are identical, only values
corresponding to the component of the fine line are set to the
difference d0 and the difference d1.
Accordingly, of the absolute values of the differences placed
corresponding to the pixels, in the event that adjacent difference
values are identical, the data continuity detecting unit 101
determines that the pixel corresponding to the absolute values of
the two differences (the pixel between the two absolute values of
difference) contains the component of the fine line.
The data continuity detecting unit 101 can also detect fine lines
with a simple method such as this.
FIG. 56 is a flowchart for describing continuity detection
processing.
In step S201, the non-continuity component extracting unit 201
extracts non-continuity component, which is portions other than the
portion where the fine line has been projected, from the input
image. The non-continuity component extracting unit 201 supplies
non-continuity component information indicating the extracted
non-continuity component, along with the input image, to the peak
detecting unit 202 and the monotonous increase/decrease detecting
unit 203. Details of the processing for extracting the
non-continuity component will be described later.
In step S202, the peak detecting unit 202 eliminates the
non-continuity component from the input image, based on the
non-continuity component information supplied from the
non-continuity component extracting unit 201, so as to leave only
pixels including the continuity component in the input image.
Further, in step S202, the peak detecting unit 202 detects
peaks.
That is to say, in the event of executing processing with the
vertical direction of the screen as a reference, of the pixels
containing the continuity component, the peak detecting unit 202
compares the pixel value of each pixel with the pixel values of the
pixels above and below, and detects pixels having a greater pixel
value than the pixel value of the pixel above and the pixel value
of the pixel below, thereby detecting a peak. Also, in step S202,
in the event of executing processing with the horizontal direction
of the screen as a reference, of the pixels containing the
continuity component, the peak detecting unit 202 compares the
pixel value of each pixel with the pixel values of the pixels to
the right side and left side, and detects pixels having a greater
pixel value than the pixel value of the pixel to the right side and
the pixel value of the pixel to the left side, thereby detecting a
peak.
The peak detecting unit 202 supplies the peak information
indicating the detected peaks to the monotonous increase/decrease
detecting unit 203.
In step S203, the monotonous increase/decrease detecting unit 203
eliminates the non-continuity component from the input image, based
on the non-continuity component information supplied from the
non-continuity component extracting unit 201, so as to leave only
pixels including the continuity component in the input image.
Further, in step S203, the monotonous increase/decrease detecting
unit 203 detects the region made up of pixels having data
continuity, by detecting monotonous increase/decrease as to the
peak, based on peak information indicating the position of the
peak, supplied from the peak detecting unit 202.
In the event of executing processing with the vertical direction of
the screen as a reference, the monotonous increase/decrease
detecting unit 203 detects monotonous increase/decrease made up of
one row of pixels aligned vertically where a single fine line image
has been projected, based on the pixel value of the peak and the
pixel values of the one row of pixels aligned vertically as to the
peak, thereby detecting a region made up of pixels having data
continuity. That is to say, in step S203, in the event of executing
processing with the vertical direction of the screen as a
reference, the monotonous increase/decrease detecting unit 203
obtains, with regard to a peak and a row of pixels aligned
vertically as to the peak, the difference between the pixel value
of each pixel and the pixel value of a pixel above or below,
thereby detecting a pixel where the sign of the difference changes.
Also, with regard to a peak and a row of pixels aligned vertically
as to the peak, the monotonous increase/decrease detecting unit 203
compares the sign of the pixel value of each pixel with the sign of
the pixel value of a pixel above or below, thereby detecting a
pixel where the sign of the pixel value changes. Further, the
monotonous increase/decrease detecting unit 203 compares pixel
value of the peak and the pixel values of the pixels to the right
side and to the left side of the peak with a threshold value, and
detects a region made up of pixels wherein the pixel value of the
peak exceeds the threshold value, and wherein the pixel values of
the pixels to the right side and to the left side of the peak are
within the threshold.
The monotonous increase/decrease detecting unit 203 takes a region
detected in this way as a monotonous increase/decrease region, and
supplies monotonous increase/decrease region information indicating
the monotonous increase/decrease region to the continuousness
detecting unit 204.
In the event of executing processing with the horizontal direction
of the screen as a reference, the monotonous increase/decrease
detecting unit 203 detects monotonous increase/decrease made up of
one row of pixels aligned horizontally where a single fine line
image has been projected, based on the pixel value of the peak and
the pixel values of the one row of pixels aligned horizontally as
to the peak, thereby detecting a region made up of pixels having
data continuity. That is to say, in step S203, in the event of
executing processing with the horizontal direction of the screen as
a reference, the monotonous increase/decrease detecting unit 203
obtains, with regard to a peak and a row of pixels aligned
horizontally as to the peak, the difference between the pixel value
of each pixel and the pixel value of a pixel to the right side or
to the left side, thereby detecting a pixel where the sign of the
difference changes. Also, with regard to a peak and a row of pixels
aligned horizontally as to the peak, the monotonous
increase/decrease detecting unit 203 compares the sign of the pixel
value of each pixel with the sign of the pixel value of a pixel to
the right side or to the left side, thereby detecting a pixel where
the sign of the pixel value changes. Further, the monotonous
increase/decrease detecting unit 203 compares pixel value of the
peak and the pixel values of the pixels to the upper side and to
the lower side of the peak with a threshold value, and detects a
region made up of pixels wherein the pixel value of the peak
exceeds the threshold value, and wherein the pixel values of the
pixels to the upper side and to the lower side of the peak are
within the threshold.
The monotonous increase/decrease detecting unit 203 takes a region
detected in this way as a monotonous increase/decrease region, and
supplies monotonous increase/decrease region information indicating
the monotonous increase/decrease region to the continuousness
detecting unit 204.
In step S204, the monotonous increase/decrease detecting unit 203
determines whether or not processing of all pixels has ended. For
example, the non-continuity component extracting unit 201 detects
peaks for all pixels of a single screen (for example, frame, field,
or the like) of the input image, and whether or not a monotonous
increase/decrease region has been detected is determined.
In the event that determination is made in step S204 that
processing of all pixels has not ended, i.e., that there are still
pixels which have not been subjected to the processing of peak
detection and detection of monotonous increase/decrease region, the
flow returns to step S202, a pixel which has not yet been subjected
to the processing of peak detection and detection of monotonous
increase/decrease region is selected as an object of the
processing, and the processing of peak detection and detection of
monotonous increase/decrease region are repeated.
In the event that determination is made in step S204 that
processing of all pixels has ended, in the event that peaks and
monotonous increase/decrease regions have been detected with regard
to all pixels, the flow proceeds to step S205, where the
continuousness detecting unit 204 detects the continuousness of
detected regions, based on the monotonous increase/decrease region
information. For example, in the event that monotonous
increase/decrease regions made up of one row of pixels aligned in
the vertical direction of the screen, indicated by monotonous
increase/decrease region information, include pixels adjacent in
the horizontal direction, the continuousness detecting unit 204
determines that there is continuousness between the two monotonous
increase/decrease regions, and in the event of not including pixels
adjacent in the horizontal direction, determines that there is no
continuousness between the two monotonous increase/decrease
regions. For example, in the event that monotonous
increase/decrease regions made up of one row of pixels aligned in
the horizontal direction of the screen, indicated by monotonous
increase/decrease region information, include pixels adjacent in
the vertical direction, the continuousness detecting unit 204
determines that there is continuousness between the two monotonous
increase/decrease regions, and in the event of not including pixels
adjacent in the vertical direction, determines that there is no
continuousness between the two monotonous increase/decrease
regions.
The continuousness detecting unit 204 takes the detected continuous
regions as continuity regions having data continuity, and outputs
data continuity information indicating the peak position and
continuity region. The data continuity information contains
information indicating the connection of regions. The data
continuity information output from the continuousness detecting
unit 204 indicates the fine line region, which is the continuity
region, made up of pixels where the actual world 1 fine line image
has been projected.
In step S206, a continuity direction detecting unit 205 determines
whether or not processing of all pixels has ended. That is to say,
the continuity direction detecting unit 205 determines whether or
not region continuation has been detected with regard to all pixels
of a certain frame of the input image.
In the event that determination is made in step S206 that
processing of all pixels has not yet ended, i.e., that there are
still pixels which have not yet been taken as the object of
detection of region continuation, the flow returns to step S205, a
pixel which has not yet been subjected to the processing of
detection of region continuity is selected, and the processing for
detection of region continuity is repeated.
In the event that determination is made in step S206 that
processing of all pixels has ended, i.e., that all pixels have been
taken as the object of detection of region continuity, the
processing ends.
Thus, the continuity contained in the data 3 which is the input
image is detected. That is to say, continuity of data included in
the data 3 which has been generated by the actual world 1 image
which is a fine line having been projected on the data 3 is
detected, and a region having data continuity, which is made up of
pixels on which the actual world 1 image which is a fine line has
been projected, is detected from the data 3.
Now, the data continuity detecting unit 101 shown in FIG. 41 can
detect time-directional data continuity, based on the region having
data continuity detected form the frame of the data 3.
For example, as shown in FIG. 57, the continuousness detecting unit
204 detects time-directional data continuity by connecting the
edges of the region having detected data continuity in frame #n,
the region having detected data continuity in frame #n-1, and the
region having detected data continuity in frame #n+1.
The frame #n-1 is a frame preceding the frame #n time-wise, and the
frame #n+1 is a frame following the frame #n time-wise. That is to
say, the frame #n-1, the frame #n, and the frame #n+1, are
displayed on the order of the frame #n-1, the frame #n, and the
frame #n+1.
More specifically, in FIG. 57, G denotes a movement vector obtained
by connecting the one edge of the region having detected data
continuity in frame #n, the region having detected data continuity
in frame #n-1, and the region having detected data continuity in
frame #n+1, and G' denotes a movement vector obtained by connecting
the other edges of the regions having detected data continuity. The
movement vector G and the movement vector G' are an example of data
continuity in the time direction.
Further, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 41 can output information indicating
the length of the region having data continuity as data continuity
information.
FIG. 58 is a block diagram illustrating the configuration of the
non-continuity component extracting unit 201 which performs planar
approximation of the non-continuity component which is the portion
of the image data which does not have data continuity, and extracts
the non-continuity component.
The non-continuity component extracting unit 201 of which the
configuration is shown in FIG. 58 extracts blocks, which are made
up of a predetermined number of pixels, from the input image,
performs planar approximation of the blocks, so that the error
between the block and a planar value is below a predetermined
threshold value, thereby extracting the non-continuity
component.
The input image is supplied to a block extracting unit 221, and is
also output without change.
The block extracting unit 221 extracts blocks, which are made up of
a predetermined number of pixels, from the input image. For
example, the block extracting unit 221 extracts a block made up of
7.times.7 pixels, and supplies this to a planar approximation unit
222. For example, the block extracting unit 221 moves the pixel
serving as the center of the block to be extracted in raster scan
order, thereby sequentially extracting blocks from the input
image.
The planar approximation unit 222 approximates the pixel values of
a pixel contained in the block on a predetermined plane. For
example, the planar approximation unit 222 approximates the pixel
value of a pixel contained in the block on a plane expressed by
Expression (24). Z=ax+by+c (24)
In Expression (24), x represents the position of the pixel in one
direction on the screen (the spatial direction X), and y represents
the position of the pixel in the other direction on the screen (the
spatial direction Y). z represents the application value
represented by the plane. a represents the gradient of the spatial
direction X of the plane, and b represents the gradient of the
spatial direction Y of the plane. In Expression (24), c represents
the offset of the plane (intercept).
For example, the planar approximation unit 222 obtains the gradient
a, gradient b, and offset c, by regression processing, thereby
approximating the pixel values of the pixels contained in the block
on a plane expressed by Expression (24). The planar approximation
unit 222 obtains the gradient a, gradient b, and offset c, by
regression processing including rejection, thereby approximating
the pixel values of the pixels contained in the block on a plane
expressed by Expression (24).
For example, the planar approximation unit 222 obtains the plane
expressed by Expression (24) wherein the error is least as to the
pixel values of the pixels of the block using the least-square
method, thereby approximating the pixel values of the pixels
contained in the block on the plane.
Note that while the planar approximation unit 222 has been
described approximating the block on the plane expressed by
Expression (24), this is not restricted to the plane expressed by
Expression (24), rather, the block may be approximated on a plane
represented with a function with a higher degree of freedom, for
example, an n-order (wherein n is an arbitrary integer)
polynomial.
A repetition determining unit 223 calculates the error between the
approximation value represented by the plane upon which the pixel
values of the block have been approximated, and the corresponding
pixel values of the pixels of the block. Expression (25) is an
expression which shows the error ei which is the difference between
the approximation value represented by the plane upon which the
pixel values of the block have been approximated, and the
corresponding pixel values zi of the pixels of the block.
.times..times. ##EQU00015##
In Expression (25), z-hat (A symbol with ^ over z will be described
as z-hat. The same description will be used in the present
specification hereafter.) represents an approximation value
expressed by the plane on which the pixel values of the block are
approximated, a-hat represents the gradient of the spatial
direction X of the plane on which the pixel values of the block are
approximated, b-hat represents the gradient of the spatial
direction Y of the plane on which the pixel values of the block are
approximated, and c-hat represents the offset (intercept) of the
plane on which the pixel values of the block are approximated.
The repetition determining unit 223 rejects the pixel regarding
which the error ei between the approximation value and the
corresponding pixel values of pixels of the block, shown in
Expression (25). Thus, pixels where the fine line has been
projected, i.e., pixels having continuity, are rejected. The
repetition determining unit 223 supplies rejection information
indicating the rejected pixels to the planar approximation unit
222.
Further, the repetition determining unit 223 calculates a standard
error, and in the event that the standard error is equal to or
greater than threshold value which has been set beforehand for
determining ending of approximation, and half or more of the pixels
of the pixels of a block have not been rejected, the repetition
determining unit 223 causes the planar approximation unit 222 to
repeat the processing of planar approximation on the pixels
contained in the block, from which the rejected pixels have been
eliminated.
Pixels having continuity are rejected, so approximating the pixels
from which the rejected pixels have been eliminated on a plane
means that the plane approximates the non-continuity component.
At the point that the standard error below the threshold value for
determining ending of approximation, or half or more of the pixels
of the pixels of a block have been rejected, the repetition
determining unit 223 ends planar approximation.
With a block made up of 5.times.5 pixels, the standard error
e.sub.s can be calculated with, for example, Expression (26).
.times..times. ##EQU00016##
Here, n is the number of pixels.
Note that the repetition determining unit 223 is not restricted to
standard error, and may be arranged to calculate the sum of the
square of errors for all of the pixels contained in the block, and
perform the following processing.
Now, at the time of planar approximation of blocks shifted one
pixel in the raster scan direction, a pixel having continuity,
indicated by the black circle in the diagram, i.e., a pixel
containing the fine line component, will be rejected multiple
times, as shown in FIG. 59.
Upon completing planar approximation, the repetition determining
unit 223 outputs information expressing the plane for approximating
the pixel values of the block (the gradient and intercept of the
plane of Expression 24)) as non-continuity information.
Note that an arrangement may be made wherein the repetition
determining unit 223 compares the number of times of rejection per
pixel with a preset threshold value, and takes a pixel which has
been rejected a number of times equal to or greater than the
threshold value as a pixel containing the continuity component, and
output the information indicating the pixel including the
continuity component as continuity component information. In this
case, the peak detecting unit 202 through the continuity direction
detecting unit 205 execute their respective processing on pixels
containing continuity component, indicated by the continuity
component information.
Examples of results of non-continuity component extracting
processing will be described with reference to FIG. 60 through FIG.
67.
FIG. 60 is a diagram illustrating an example of an input image
generated by the average value of the pixel values of 2.times.2
pixels in an original image containing fine lines having been
generated as a pixel value.
FIG. 61 is a diagram illustrating an image from the image shown in
FIG. 60 wherein standard error obtained as the result of planar
approximation without rejection is taken as the pixel value. In the
example shown in FIG. 61, a block made up of 5.times.5 pixels as to
a single pixel of interest was subjected to planar approximation.
In FIG. 61, white pixels are pixel values which have greater pixel
values, i.e., pixels having greater standard error, and black
pixels are pixel values which have smaller pixel values, i.e.,
pixels having smaller standard error.
From FIG. 61, it can be confirmed that in the event that the
standard error obtained as the result of planar approximation
without rejection is taken as the pixel value, great values are
obtained over a wide area at the perimeter of non-continuity
portions.
In the examples shown in FIG. 62 through FIG. 67, a block made up
of 7.times.7 pixels as to a single pixel of interest was subjected
to planar approximation. In the event of planar approximation of a
block made up of 7.times.7 pixels, one pixel is repeatedly included
in 49 blocks, meaning that a pixel containing the continuity
component is rejected as many as 49 times.
FIG. 62 is an image wherein standard error obtained by planar
approximation with rejection of the image shown in FIG. 60 is taken
as the pixel value.
In FIG. 62, white pixels are pixel values which have greater pixel
values, i.e., pixels having greater standard error, and black
pixels are pixel values which have smaller pixel values, i.e.,
pixels having smaller standard error. It can be understood that the
standard error is smaller overall in the case of performing
rejection, as compared with a case of not performing rejection.
FIG. 63 is an image wherein the number of times of rejection in
planar approximation with rejection of the image shown in FIG. 60
is taken as the pixel value. In FIG. 63, white pixels are greater
pixel values, i.e., pixels which have been rejected a greater
number of times, and black pixels are smaller pixel values, i.e.,
pixels which have been rejected a fewer times.
From FIG. 63, it can be understood that pixels where the fine line
images are projected have been discarded a greater number of times.
An image for masking the non-continuity portions of the input image
can be generated using the image wherein the number of times of
rejection is taken as the pixel value.
FIG. 64 is a diagram illustrating an image wherein the gradient of
the spatial direction X of the plane for approximating the pixel
values of the block is taken as the pixel value. FIG. 65 is a
diagram illustrating an image wherein the gradient of the spatial
direction Y of the plane for approximating the pixel values of the
block is taken as the pixel value.
FIG. 66 is a diagram illustrating an image formed of approximation
values expressed by a plane for approximating the pixel values of
the block. It can be understood that the fine lines have
disappeared from the image shown in FIG. 66.
FIG. 67 is a diagram illustrating an image made up of the
difference between the image shown in FIG. 60 generated by the
average value of the block of 2.times.2 pixels in the original
image being taken as the pixel value, and an image made up of
approximate values expressed as a plane, shown in FIG. 66. The
pixel values of the image shown in FIG. 67 have had the
non-continuity component removed, so only the values where the
image of the fine line has been projected remain. As can be
understood from FIG. 67, with an image made up of the difference
between the pixel value of the original image and approximation
values expressed by a plane whereby approximation has been
performed, the continuity component of the original image is
extracted well.
The number of times of rejection, the gradient of the spatial
direction X of the plane for approximating the pixel values of the
pixel of the block, the gradient of the spatial direction Y of the
plane for approximating the pixel values of the pixel of the block,
approximation values expressed by the plane approximating the pixel
values of the pixels of the block, and the error ei, can be used as
features of the input image.
FIG. 68 is a flowchart for describing the processing of extracting
the non-continuity component with the non-continuity component
extracting unit 201 of which the configuration is shown in FIG.
58.
In step S221, the block extracting unit 221 extracts a block made
up of a predetermined number of pixels from the input image, and
supplies the extracted block to the planar approximation unit 222.
For example, the block extracting unit 221 selects one pixel of the
pixels of the input pixel which have not been selected yet, and
extracts a block made up of 7.times.7 pixels centered on the
selected pixel. For example, the block extracting unit 221 can
select pixels in raster scan order.
In step S222, the planar approximation unit 222 approximates the
extracted block on a plane. The planar approximation unit 222
approximates the pixel values of the pixels of the extracted block
on a plane by regression processing, for example. For example, the
planar approximation unit 222 approximates the pixel values of the
pixels of the extracted block excluding the rejected pixels on a
plane, by regression processing. In step S223, the repetition
determining unit 223 executes repetition determination. For
example, repetition determination is performed by calculating the
standard error from the pixel values of the pixels of the block and
the planar approximation values, and counting the number of
rejected pixels.
In step S224, the repetition determining unit 223 determines
whether or not the standard error is equal to or above a threshold
value, and in the event that determination is made that the
standard error is equal to or above the threshold value, the flow
proceeds to step S225.
Note that an arrangement may be made wherein the repetition
determining unit 223 determines in step S224 whether or not half or
more of the pixels of the block have been rejected, and whether or
not the standard error is equal to or above the threshold value,
and in the event that determination is made that half or more of
the pixels of the block have not been rejected, and the standard
error is equal to or above the threshold value, the flow proceeds
to step S225.
In step S225, the repetition determining unit 223 calculates the
error between the pixel value of each pixel of the block and the
approximated planar approximation value, rejects the pixel with the
greatest error, and notifies the planar approximation unit 222. The
procedure returns to step S222, and the planar approximation
processing and repetition determination processing is repeated with
regard to the pixels of the block excluding the rejected pixel.
In step S225, in the event that a block which is shifted one pixel
in the raster scan direction is extracted in the processing in step
S221, the pixel including the fine line component (indicated by the
black circle in the drawing) is rejected multiple times, as shown
in FIG. 59.
In the event that determination is made in step S224 that the
standard error is not equal to or greater than the threshold value,
the block has been approximated on the plane, so the flow proceeds
to step S226.
Note that an arrangement may be made wherein the repetition
determining unit 223 determines in step S224 whether or not half or
more of the pixels of the block have been rejected, and whether or
not the standard error is equal to or above the threshold value,
and in the event that determination is made that half or more of
the pixels of the block have been rejected, or the standard error
is not equal to or above the threshold value, the flow proceeds to
step S225.
In step S226, the repetition determining unit 223 outputs the
gradient and intercept of the plane for approximating the pixel
values of the pixels of the block as non-continuity component
information.
In step S227, the block extracting unit 221 determines whether or
not processing of all pixels of one screen of the input image has
ended, and in the event that determination is made that there are
still pixels which have not yet been taken as the object of
processing, the flow returns to step S221, a block is extracted
from pixels not yet been subjected to the processing, and the above
processing is repeated.
In the event that determination is made in step S227 that
processing has ended for all pixels of one screen of the input
image, the processing ends.
Thus, the non-continuity component extracting unit 201 of which the
configuration is shown in FIG. 58 can extract the non-continuity
component from the input image. The non-continuity component
extracting unit 201 extracts the non-continuity component from the
input image, so the peak detecting unit 202 and monotonous
increase/decrease detecting unit 203 can obtain the difference
between the input image and the non-continuity component extracted
by the non-continuity component extracting unit 201, so as to
execute the processing regarding the difference containing the
continuity component.
Note that the standard error in the event that rejection is
performed, the standard error in the event that rejection is not
performed, the number of times of rejection of a pixel, the
gradient of the spatial direction X of the plane (a-hat in
Expression (24)), the gradient of the spatial direction Y of the
plane (b-hat in Expression (24)), the level of planar transposing
(c-hat in Expression (24)), and the difference between the pixel
values of the input image and the approximation values represented
by the plane, calculated in planar approximation processing, can be
used as features.
FIG. 69 is a flowchart for describing processing for extracting the
continuity component with the non-continuity component extracting
unit 201 of which the configuration is shown in FIG. 58, instead of
the processing for extracting the non-continuity component
corresponding to step S201. The processing of step S241 through
step S245 is the same as the processing of step S221 through step
S225, so description thereof will be omitted.
In step S246, the repetition determining unit 223 outputs the
difference between the approximation value represented by the plane
and the pixel values of the input image, as the continuity
component of the input image. That is to say, the repetition
determining unit 223 outputs the difference between the planar
approximation values and the true pixel values.
Note that the repetition determining unit 223 may be arranged to
output the difference between the approximation value represented
by the plane and the pixel values of the input image, regarding
pixel values of pixels of which the difference is equal to or
greater than a predetermined threshold value, as the continuity
component of the input image.
The processing of step S247 is the same as the processing of step
S227, and accordingly description thereof will be omitted.
The plane approximates the non-continuity component, so the
non-continuity component extracting unit 201 can remove the
non-continuity component from the input image by subtracting the
approximation value represented by the plane for approximating
pixel values, from the pixel values of each pixel in the input
image. In this case, the peak detecting unit 202 through the
continuousness detecting unit 204 can be made to process only the
continuity component of the input image, i.e., the values where the
fine line image has been projected, so the processing with the peak
detecting unit 202 through the continuousness detecting unit 204
becomes easier.
FIG. 70 is a flowchart for describing other processing for
extracting the continuity component with the non-continuity
component extracting unit 201 of which the configuration is shown
in FIG. 58, instead of the processing for extracting the
non-continuity component corresponding to step S201. The processing
of step S261 through step S265 is the same as the processing of
step S221 through step S225, so description thereof will be
omitted.
In step S266, the repetition determining unit 223 stores the number
of times of rejection for each pixel, the flow returns to step
S262, and the processing is repeated.
In step S264, in the event that determination is made that the
standard error is not equal to or greater than the threshold value,
the block has been approximated on the plane, so the flow proceeds
to step S267, the repetition determining unit 223 determines
whether or not processing of all pixels of one screen of the input
image has ended, and in the event that determination is made that
there are still pixels which have not yet been taken as the object
of processing, the flow returns to step S261, with regard to a
pixel which has not yet been subjected to the processing, a block
is extracted, and the above processing is repeated.
In the event that determination is made in step S267 that
processing has ended for all pixels of one screen of the input
image, the flow proceeds to step S268, the repetition determining
unit 223 selects a pixel which has not yet been selected, and
determines whether or not the number of times of rejection of the
selected pixel is equal to or greater than a threshold value. For
example, the repetition determining unit 223 determines in step
S268 whether or not the number of times of rejection of the
selected pixel is equal to or greater than a threshold value stored
beforehand.
In the event that determination is made in step S268 that the
number of times of rejection of the selected pixel is equal to or
greater than the threshold value, the selected pixel contains the
continuity component, so the flow proceeds to step S269, where the
repetition determining unit 223 outputs the pixel value of the
selected pixel (the pixel value in the input image) as the
continuity component of the input image, and the flow proceeds to
step S270.
In the event that determination is made in step S268 that the
number of times of rejection of the selected pixel is not equal to
or greater than the threshold value, the selected pixel does not
contain the continuity component, so the processing in step S269 is
skipped, and the procedure proceeds to step S270. That is to say,
the pixel value of a pixel regarding which determination has been
made that the number of times of rejection is not equal to or
greater than the threshold value is not output.
Note that an arrangement may be made wherein the repetition
determining unit 223 outputs a pixel value set to 0 for pixels
regarding which determination has been made that the number of
times of rejection is not equal to or greater than the threshold
value.
In step S270, the repetition determining unit 223 determines
whether or not processing of all pixels of one screen of the input
image has ended to determine whether or not the number of times of
rejection is equal to or greater than the threshold value, and in
the event that determination is made that processing has not ended
for all pixels, this means that there are still pixels which have
not yet been taken as the object of processing, so the flow returns
to step S268, a pixel which has not yet been subjected to the
processing is selected, and the above processing is repeated.
In the event that determination is made in step S270 that
processing has ended for all pixels of one screen of the input
image, the processing ends.
Thus, of the pixels of the input image, the non-continuity
component extracting unit 201 can output the pixel values of pixels
containing the continuity component, as continuity component
information. That is to say, of the pixels of the input image, the
non-continuity component extracting unit 201 can output the pixel
values of pixels containing the component of the fine line
image.
FIG. 71 is a flowchart for describing yet other processing for
extracting the continuity component with the non-continuity
component extracting unit 201 of which the configuration is shown
in FIG. 58, instead of the processing for extracting the
non-continuity component corresponding to step S201. The processing
of step S281 through step S288 is the same as the processing of
step S261 through step S268, so description thereof will be
omitted.
In step S289, the repetition determining unit 223 outputs the
difference between the approximation value represented by the
plane, and the pixel value of a selected pixel, as the continuity
component of the input image. That is to say, the repetition
determining unit 223 outputs an image wherein the non-continuity
component has been removed from the input image, as the continuity
information.
The processing of step S290 is the same as the processing of step
S270, and accordingly description thereof will be omitted.
Thus, the non-continuity component extracting unit 201 can output
an image wherein the non-continuity component has been removed from
the input image as the continuity information.
As described above, in a case wherein real world light signals are
projected, a non-continuous portion of pixel values of multiple
pixels of first image data wherein a part of the continuity of the
real world light signals has been lost is detected, data continuity
is detected from the detected non-continuous portions, a model
(function) is generated for approximating the light signals by
estimating the continuity of the real world light signals based on
the detected data continuity, and second image data is generated
based on the generated function, processing results which are more
accurate and have higher precision as to the event in the real
world can be obtained.
FIG. 72 is a block diagram illustrating another configuration of
the data continuity detecting unit 101.
With the data continuity detecting unit 101 of which the
configuration is shown in FIG. 72, change in the pixel value of the
pixel of interest which is a pixel of interest in the spatial
direction of the input image, i.e. activity in the spatial
direction of the input image, is detected, multiple sets of pixels
made up of a predetermined number of pixels in one row in the
vertical direction or one row in the horizontal direction are
extracted for each angle based on the pixel of interest and a
reference axis according to the detected activity, the correlation
of the extracted pixel sets is detected, and the angle of data
continuity based on the reference axis in the input image is
detected based on the correlation.
The angle of data continuity means an angle assumed by the
reference axis, and the direction of a predetermined dimension
where constant characteristics repeatedly appear in the data 3.
Constant characteristics repeatedly appearing means a case wherein,
for example, the change in value as to the change in position in
the data 3, i.e., the cross-sectional shape, is the same, and so
forth.
The reference axis may be, for example, an axis indicating the
spatial direction X (the horizontal direction of the screen), an
axis indicating the spatial direction Y (the vertical direction of
the screen), and so forth.
The input image is supplied to an activity detecting unit 401 and
data selecting unit 402.
The activity detecting unit 401 detects change in the pixel values
as to the spatial direction of the input image, i.e., activity in
the spatial direction, and supplies the activity information which
indicates the detected results to the data selecting unit 402 and a
continuity direction derivation unit 404.
For example, the activity detecting unit 401 detects the change of
a pixel value as to the horizontal direction of the screen, and the
change of a pixel value as to the vertical direction of the screen,
and compares the detected change of the pixel value in the
horizontal direction and the change of the pixel value in the
vertical direction, thereby detecting whether the change of the
pixel value in the horizontal direction is greater as compared with
the change of the pixel value in the vertical direction, or whether
the change of the pixel value in the vertical direction is greater
as compared with the change of the pixel value in the horizontal
direction.
The activity detecting unit 401 supplies to the data selecting unit
402 and the continuity direction derivation unit 404 activity
information, which is the detection results, indicating that the
change of the pixel value in the horizontal direction is greater as
compared with the change of the pixel value in the vertical
direction, or indicating that the change of the pixel value in the
vertical direction is greater as compared with the change of the
pixel value in the horizontal direction.
In the event that the change of the pixel value in the horizontal
direction is greater as compared with the change of the pixel value
in the vertical direction, arc shapes (half-disc shapes) or pawl
shapes are formed on one row in the vertical direction, as
indicated by FIG. 73 for example, and the arc shapes or pawl shapes
are formed repetitively more in the vertical direction. That is to
say, in the event that the change of the pixel value in the
horizontal direction is greater as compared with the change of the
pixel value in the vertical direction, with the reference axis as
the axis representing the spatial direction X, the angle of the
data continuity based on the reference axis in the input image is a
value of any from 45 degrees to 90 degrees.
In the event that the change of the pixel value in the vertical
direction is greater as compared with the change of the pixel value
in the horizontal direction, arc shapes or pawl shapes are formed
on one row in the vertical direction, for example, and the arc
shapes or pawl shapes are formed repetitively more in the
horizontal direction. That is to say, in the event that the change
of the pixel value in the vertical direction is greater as compared
with the change of the pixel value in the horizontal direction,
with the reference axis as the axis representing the spatial
direction X, the angle of the data continuity based on the
reference axis in the input image is a value of any from 0 degrees
to 45 degrees.
For example, the activity detecting unit 401 extracts from the
input image a block made up of the 9 pixels, 3.times.3 centered on
the pixel of interest, as shown in FIG. 74. The activity detecting
unit 401 calculates the sum of differences of the pixels values
regarding the pixels vertically adjacent, and the sum of
differences of the pixels values regarding the pixels horizontally
adjacent. The sum of differences h.sub.diff of the pixels values
regarding the pixels horizontally adjacent can be obtained with
Expression (27). h.sub.diff=.SIGMA.(P.sub.i+1,j-P.sub.i,j) (27)
In the same way, the sum of differences v.sub.diff of the pixels
values regarding the pixels vertically adjacent can be obtained
with Expression (28). v.sub.diff=.SIGMA.(P.sub.i,j+1-P.sub.i,j)
(28)
In Expression (27) and Expression (28), P represents the pixel
value, i represents the position of the pixel in the horizontal
direction, and j represents the position of the pixel in the
vertical direction.
An arrangement may be made wherein the activity detecting unit 401
compares the calculated sum of differences h.sub.diff of the pixels
values regarding the pixels horizontally adjacent with the sum of
differences v.sub.diff of the pixels values regarding the pixels
vertically adjacent, so as to determine the range of the angle of
the data continuity based on the reference axis in the input image.
That is to say, in this case, the activity detecting unit 401
determines whether a shape indicated by change in the pixel value
as to the position in the spatial direction is formed repeatedly in
the horizontal direction, or formed repeatedly in the vertical
direction.
For example, change in pixel values in the horizontal direction
with regard to an arc formed on pixels in one horizontal row is
greater than the change of pixel values in the vertical direction,
change in pixel values in the vertical direction with regard to an
arc formed on pixels in one horizontal row is greater than the
change of pixel values in the horizontal direction, and it can be
said that the direction of data continuity, i.e., the change in the
direction of the predetermined dimension of a constant feature
which the input image that is the data 3 has is smaller in
comparison with the change in the orthogonal direction too the data
continuity. In other words, the difference of the direction
orthogonal to the direction of data continuity (hereafter also
referred to as non-continuity direction) is greater as compared to
the difference in the direction of data continuity.
For example, as shown in FIG. 75, the activity detecting unit 401
compares the calculated sum of differences h.sub.diff of the pixels
values regarding the pixels horizontally adjacent with the sum of
differences v.sub.diff of the pixels values regarding the pixels
vertically adjacent, and in the event that the sum of differences
h.sub.diff of the pixels values regarding the pixels horizontally
adjacent is greater, determines that the angle of the data
continuity based on the reference axis is a value of any from 45
degrees to 135 degrees, and in the event that the sum of
differences v.sub.diff of the pixels values regarding the pixels
vertically adjacent is greater, determines that the angle of the
data continuity based on the reference axis is a value of any from
0 degrees to 45 degrees, or a value of any from 135 degrees to 180
degrees.
For example, the activity detecting unit 401 supplies activity
information indicating the determination results to the data
selecting unit 402 and the continuity direction derivation unit
404.
Note that the activity detecting unit 401 can detect activity by
extracting blocks of arbitrary sizes, such as a block made up of 25
pixels of 5.times.5, a block made up of 49 pixels of 7.times.7, and
so forth.
The data selecting unit 402 sequentially selects pixels of interest
from the pixels of the input image, and extracts multiple sets of
pixels made up of a predetermined number of pixels in one row in
the vertical direction or one row in the horizontal direction for
each angle based on the pixel of interest and the reference axis,
based on the activity information supplied from the activity
detecting unit 401.
For example, in the event that the activity information indicates
that the change in pixel values in the horizontal direction is
greater in comparison with the change in pixel values in the
vertical direction, this means that the data continuity angle is a
value of any from 45 degrees to 135 degrees, so the data selecting
unit 402 extracts multiple sets of pixels made up of a
predetermined number of pixels in one row in the vertical
direction, for each predetermined angle in the range of 45 degrees
to 135 degrees, based on the pixel of interest and the reference
axis.
In the event that the activity information indicates that the
change in pixel values in the vertical direction is greater in
comparison with the change in pixel values in the horizontal
direction, this means that the data continuity angle is a value of
any from 0 degrees to 45 degrees or from 135 degrees to 180
degrees, so the data selecting unit 402 extracts multiple sets of
pixels made up of a predetermined number of pixels in one row in
the horizontal direction, for each predetermined angle in the range
of 0 degrees to 45 degrees or 135 degrees to 180 degrees, based on
the pixel of interest and the reference axis.
Also, for example, in the event that the activity information
indicates that the angle of data continuity is a value of any from
45 degrees to 135 degrees, the data selecting unit 402 extracts
multiple sets of pixels made up of a predetermined number of pixels
in one row in the vertical direction, for each predetermined angle
in the range of 45 degrees to 135 degrees, based on the pixel of
interest and the reference axis.
In the event that the activity information indicates that the angle
of data continuity is a value of any from 0 degrees to 45 degrees
or from 135 degrees to 180 degrees, the data selecting unit 402
extracts multiple sets of pixels made up of a predetermined number
of pixels in one row in the horizontal direction, for each
predetermined angle in the range of 0 degrees to 45 degrees or 135
degrees to 180 degrees, based on the pixel of interest and the
reference axis.
The data selecting unit 402 supplies the multiple sets made up of
the extracted pixels to an error estimating unit 403.
The error estimating unit 403 detects correlation of pixel sets for
each angle with regard to the multiple sets of extracted
pixels.
For example, with regard to the multiple sets of pixels made up of
a predetermined number of pixels in one row in the vertical
direction corresponding to one angle, the error estimating unit 403
detects the correlation of the pixels values of the pixels at
corresponding positions of the pixel sets. With regard to the
multiple sets of pixels made up of a predetermined number of pixels
in one row in the horizontal direction corresponding to one angle,
the error estimating unit 403 detects the correlation of the pixels
values of the pixels at corresponding positions of the sets.
The error estimating unit 403 supplies correlation information
indicating the detected correlation to the continuity direction
derivation unit 404. The error estimating unit 403 calculates the
sum of the pixel values of pixels of a set including the pixel of
interest supplied from the data selecting unit 402 as values
indicating correlation, and the absolute value of difference of the
pixel values of the pixels at corresponding positions in other
sets, and supplies the sum of absolute value of difference to the
continuity direction derivation unit 404 as correlation
information.
Based on the correlation information supplied from the error
estimating unit 403, the continuity direction derivation unit 404
detects the data continuity angle based on the reference axis in
the input image, corresponding to the lost continuity of the light
signals of the actual world 1, and outputs data continuity
information indicating an angle. For example, based on the
correlation information supplied from the error estimating unit
403, the continuity direction derivation unit 404 detects an angle
corresponding to the pixel set with the greatest correlation as the
data continuity angle, and outputs data continuity information
indicating the angle corresponding to the pixel set with the
greatest correlation that has been detected.
The following description will be made regarding detection of data
continuity angle in the range of 0 degrees through 90 degrees (the
so-called first quadrant).
FIG. 76 is a block diagram illustrating a more detailed
configuration of the data continuity detecting unit 101 shown in
FIG. 72.
The data selecting unit 402 includes pixel selecting unit 411-1
through pixel selecting unit 411-L. The error estimating unit 403
includes estimated error calculating unit 412-1 through estimated
error calculating unit 412-L. The continuity direction derivation
unit 404 includes a smallest error angle selecting unit 413.
First, description will be made regarding the processing of the
pixel selecting unit 411-1 through pixel selecting unit 411-L in
the event that the data continuity angle indicated by the activity
information is a value of any from 45 degrees to 135 degrees.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
set straight lines of mutually differing predetermined angles which
pass through the pixel of interest, with the axis indicating the
spatial direction X as the reference axis. The pixel selecting unit
411-1 through pixel selecting unit 411-L select, of the pixels
belonging to a vertical row of pixels to which the pixel of
interest belongs, a predetermined number of pixels above the pixel
of interest, and predetermined number of pixels below the pixel of
interest, and the pixel of interest, as a set.
For example, as shown in FIG. 77, the pixel selecting unit 411-1
through pixel selecting unit 411-L select 9 pixels centered on the
pixel of interest, as a set of pixels, from the pixels belonging to
a vertical row of pixels to which the pixel of interest
belongs.
In FIG. 77, one grid-shaped square (one grid) represents one pixel.
In FIG. 77, the circle shown at the center represents the pixel of
interest.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a vertical row of pixels to the
left of the vertical row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. In FIG. 77, the circle to the lower left of the pixel of
interest represents an example of a selected pixel. The pixel
selecting unit 411-1 through pixel selecting unit 411-L then
select, from the pixels belonging to the vertical row of pixels to
the left of the vertical row of pixels to which the pixel of
interest belongs, a predetermined number of pixels above the
selected pixel, a predetermined number of pixels below the selected
pixel, and the selected pixel, as a set of pixels.
For example, as shown in FIG. 77, the pixel selecting unit 411-1
through pixel selecting unit 411-L select 9 pixels centered on the
pixel at the position closest to the straight line, from the pixels
belonging to the vertical row of pixels to the left of the vertical
row of pixels to which the pixel of interest belongs, as a set of
pixels.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a vertical row of pixels second
left from the vertical row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. In FIG. 77, the circle to the far left represents an
example of the selected pixel. The pixel selecting unit 411-1
through pixel selecting unit 411-L then select, as a set of pixels,
from the pixels belonging to the vertical row of pixels second left
from the vertical row of pixels to which the pixel of interest
belongs, a predetermined number of pixels above the selected pixel,
a predetermined number of pixels below the selected pixel, and the
selected pixel.
For example, as shown in FIG. 77, the pixel selecting unit 411-1
through pixel selecting unit 411-L select 9 pixels centered on the
pixel at the position closest to the straight line, from the pixels
belonging to the vertical row of pixels second left from the
vertical row of pixels to which the pixel of interest belongs, as a
set of pixels.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a vertical row of pixels to the
right of the vertical row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. In FIG. 77, the circle to the upper right of the pixel of
interest represents an example of a selected pixel. The pixel
selecting unit 411-1 through pixel selecting unit 411-L then
select, from the pixels belonging to the vertical row of pixels to
the right of the vertical row of pixels to which the pixel of
interest belongs, a predetermined number of pixels above the
selected pixel, a predetermined number of pixels below the selected
pixel, and the selected pixel, as a set of pixels.
For example, as shown in FIG. 77, the pixel selecting unit 411-1
through pixel selecting unit 411-L select 9 pixels centered on the
pixel at the position closest to the straight line, from the pixels
belonging to the vertical row of pixels to the right of the
vertical row of pixels to which the pixel of interest belongs, as a
set of pixels.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a vertical row of pixels second
right from the vertical row of pixels to which the pixel of
interest belongs, a pixel at the position closest to the straight
line set for each. In FIG. 77, the circle to the far right
represents an example of the selected pixel. The pixel selecting
unit 411-1 through pixel selecting unit 411-L then select, from the
pixels belonging to the vertical row of pixels second right from
the vertical row of pixels to which the pixel of interest belongs,
a predetermined number of pixels above the selected pixel, a
predetermined number of pixels below the selected pixel, and the
selected pixel, as a set of pixels.
For example, as shown in FIG. 77, the pixel selecting unit 411-1
through pixel selecting unit 411-L select 9 pixels centered on the
pixel at the position closest to the straight line, from the pixels
belonging to the vertical row of pixels second right from the
vertical row of pixels to which the pixel of interest belongs, as a
set of pixels.
Thus, the pixel selecting unit 411-1 through pixel selecting unit
411-L each select five sets of pixels.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select pixel sets for (lines set to) mutually different angles. For
example, the pixel selecting unit 411-1 selects sets of pixels
regarding 45 degrees, the pixel selecting unit 411-2 selects sets
of pixels regarding 47.5 degrees, and the pixel selecting unit
411-3 selects sets of pixels regarding 50 degrees. The pixel
selecting unit 411-1 through pixel selecting unit 411-L select sets
of pixels regarding angles every 2.5 degrees, from 52.5 degrees
through 135 degrees.
Note that the number of pixel sets may be an optional number, such
as 3 or 7, for example, and does not restrict the present
invention. Also, the number of pixels selected as one set may be an
optional number, such as 5 or 13, for example, and does not
restrict the present invention.
Note that the pixel selecting unit 411-1 through pixel selecting
unit 411-L may be arranged to select pixel sets from pixels within
a predetermined range in the vertical direction. For example, the
pixel selecting unit 411-1 through pixel selecting unit 411-L can
select pixel sets from 121 pixels in the vertical direction (60
pixels upward from the pixel of interest, and 60 pixels downward).
In this case, the data continuity detecting unit 101 can detect the
angle of data continuity up to 88.09 degrees as to the axis
representing the spatial direction X.
The pixel selecting unit 411-1 supplies the selected set of pixels
to the estimated error calculating unit 412-1, and the pixel
selecting unit 411-2 supplies the selected set of pixels to the
estimated error calculating unit 412-2. In the same way, each pixel
selecting unit 411-3 through pixel selecting unit 411-L supplies
the selected set of pixels to each estimated error calculating unit
412-3 through estimated error calculating unit 412-L.
The estimated error calculating unit 412-1 through estimated error
calculating unit 412-L detect the correlation of the pixels values
of the pixels at positions in the multiple sets, supplied from each
of the pixel selecting unit 411-1 through pixel selecting unit
411-L. For example, the estimated error calculating unit 412-1
through estimated error calculating unit 412-L calculates, as a
value indicating the correlation, the sum of absolute values of
difference between the pixel values of the pixels of the set
containing the pixel of interest, and the pixel values of the
pixels at corresponding positions in other sets, supplied from one
of the pixel selecting unit 411-1 through pixel selecting unit
411-L.
More specifically, based on the pixel values of the pixels of the
set containing the pixel of interest and the pixel values of the
pixels of the set made up of pixels belonging to one vertical row
of pixels to the left side of the pixel of interest supplied from
one of the pixel selecting unit 411-1 through pixel selecting unit
411-L, the estimated error calculating unit 412-1 through estimated
error calculating unit 412-L calculates the difference of the pixel
values of the topmost pixel, then calculates the difference of the
pixel values of the second pixel from the top, and so on to
calculate the absolute values of difference of the pixel values in
order from the top pixel, and further calculates the sum of
absolute values of the calculated differences. Based on the pixel
values of the pixels of the set containing the pixel of interest
and the pixel values of the pixels of the set made up of pixels
belonging to one vertical row of pixels two to the left from the
pixel of interest supplied from one of the pixel selecting unit
411-1 through pixel selecting unit 411-L, the estimated error
calculating unit 412-1 through estimated error calculating unit
412-L calculates the absolute values of difference of the pixel
values in order from the top pixel, and calculates the sum of
absolute values of the calculated differences.
Then, based on the pixel values of the pixels of the set containing
the pixel of interest and the pixel values of the pixels of the set
made up of pixels belonging to one vertical row of pixels to the
right side of the pixel of interest supplied from one of the pixel
selecting unit 411-1 through pixel selecting unit 411-L, the
estimated error calculating unit 412-1 through estimated error
calculating unit 412-L calculates the difference of the pixel
values of the topmost pixel, then calculates the difference of the
pixel values of the second pixel from the top, and so on to
calculate the absolute values of difference of the pixel values in
order from the top pixel, and further calculates the sum of
absolute values of the calculated differences. Based on the pixel
values of the pixels of the set containing the pixel of interest
and the pixel values of the pixels of the set made up of pixels
belonging to one vertical row of pixels two to the right from the
pixel of interest supplied from one of the pixel selecting unit
411-1 through pixel selecting unit 411-L, the estimated error
calculating unit 412-1 through estimated error calculating unit
412-L calculates the absolute values of difference of the pixel
values in order from the top pixel, and calculates the sum of
absolute values of the calculated differences.
The estimated error calculating unit 412-1 through estimated error
calculating unit 412-L add all of the sums of absolute values of
difference of the pixel values thus calculated, thereby calculating
the aggregate of absolute values of difference of the pixel
values.
The estimated error calculating unit 412-1 through estimated error
calculating unit 412-L supply information indicating the detected
correlation to the smallest error angle selecting unit 413. For
example, the estimated error calculating unit 412-1 through
estimated error calculating unit 412-L supply the aggregate of
absolute values of difference of the pixel values calculated, to
the smallest error angle selecting unit 413.
Note that the estimated error calculating unit 412-1 through
estimated error calculating unit 412-L are not restricted to the
sum of absolute values of difference of pixel values, and can also
calculate other values as correlation values as well, such as the
sum of squared differences of pixel values, or correlation
coefficients based on pixel values, and so forth.
The smallest error angle selecting unit 413 detects the data
continuity angle based on the reference axis in the input image
which corresponds to the continuity of the image which is the lost
actual world 1 light signals, based on the correlation detected by
the estimated error calculating unit 412-1 through estimated error
calculating unit 412-L with regard to mutually different angles.
That is to say, based on the correlation detected by the estimated
error calculating unit 412-1 through estimated error calculating
unit 412-L with regard to mutually different angles, the smallest
error angle selecting unit 413 selects the greatest correlation,
and takes the angle regarding which the selected correlation was
detected as the data continuity angle based on the reference axis,
thereby detecting the data continuity angle based on the reference
axis in the input image.
For example, of the aggregates of absolute values of difference of
the pixel values supplied from the estimated error calculating unit
412-1 through estimated error calculating unit 412-L, the smallest
error angle selecting unit 413 selects the smallest aggregate. With
regard to the pixel set of which the selected aggregate was
calculated, the smallest error angle selecting unit 413 makes
reference to a pixel belonging to the one vertical row of pixels
two to the left from the pixel of interest and at the closest
position to the straight line, and to a pixel belonging to the one
vertical row of pixels two to the right from the pixel of interest
and at the closest position to the straight line.
As shown in FIG. 77, the smallest error angle selecting unit 413
obtains the distance S in the vertical direction of the position of
the pixels to reference from the position of the pixel of interest.
As shown in FIG. 78, the smallest error angle selecting unit 413
calculates the angle .theta. of data continuity based on the axis
indicating the spatial direction X which is the reference axis in
the input image which is image data, that corresponds to the lost
actual world 1 light signals continuity, from Expression (29).
.theta.=tan.sup.-1(s/2) (29)
Next, description will be made regarding the processing of the
pixel selecting unit 411-1 through pixel selecting unit 411-L in
the event that the data continuity angle indicated by the activity
information is a value of any from 0 degrees to 45 degrees and 135
degrees to 180 degrees.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
set straight lines of predetermined angles which pass through the
pixel of interest, with the axis indicating the spatial direction X
as the reference axis, and select, of the pixels belonging to a
horizontal row of pixels to which the pixel of interest belongs, a
predetermined number of pixels to the left of the pixel of
interest, and predetermined number of pixels to the right of the
pixel of interest, and the pixel of interest, as a pixel set.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a horizontal row of pixels above
the horizontal row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. The pixel selecting unit 411-1 through pixel selecting
unit 411-L then select, from the pixels belonging to the horizontal
row of pixels above the horizontal row of pixels to which the pixel
of interest belongs, a predetermined number of pixels to the left
of the selected pixel, a predetermined number of pixels to the
right of the selected pixel, and the selected pixel, as a pixel
set.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a horizontal row of pixels two
above the horizontal row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. The pixel selecting unit 411-1 through pixel selecting
unit 411-L then select, from the pixels belonging to the horizontal
row of pixels two above the horizontal row of pixels to which the
pixel of interest belongs, a predetermined number of pixels to the
left of the selected pixel, a predetermined number of pixels to the
right of the selected pixel, and the selected pixel, as a pixel
set.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a horizontal row of pixels below
the horizontal row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. The pixel selecting unit 411-1 through pixel selecting
unit 411-L then select, from the pixels belonging to the horizontal
row of pixels below the horizontal row of pixels to which the pixel
of interest belongs, a predetermined number of pixels to the left
of the selected pixel, a predetermined number of pixels to the
right of the selected pixel, and the selected pixel, as a pixel
set.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select, from pixels belonging to a horizontal row of pixels two
below the horizontal row of pixels to which the pixel of interest
belongs, a pixel at the position closest to the straight line set
for each. The pixel selecting unit 411-1 through pixel selecting
unit 411-L then select, from the pixels belonging to the horizontal
row of pixels two below the horizontal row of pixels to which the
pixel of interest belongs, a predetermined number of pixels to the
left of the selected pixel, a predetermined number of pixels to the
right of the selected pixel, and the selected pixel, as a pixel
set.
Thus, the pixel selecting unit 411-1 through pixel selecting unit
411-L each select five sets of pixels.
The pixel selecting unit 411-1 through pixel selecting unit 411-L
select pixel sets for mutually different angles. For example, the
pixel selecting unit 411-1 selects sets of pixels regarding 0
degrees, the pixel selecting unit 411-2 selects sets of pixels
regarding 2.5 degrees, and the pixel selecting unit 411-3 selects
sets of pixels regarding 5 degrees. The pixel selecting unit 411-1
through pixel selecting unit 411-L select sets of pixels regarding
angles every 2.5 degrees, from 7.5 degrees through 45 degrees and
from 135 degrees through 180 degrees.
The pixel selecting unit 411-1 supplies the selected set of pixels
to the estimated error calculating unit 412-1, and the pixel
selecting unit 411-2 supplies the selected set of pixels to the
estimated error calculating unit 412-2. In the same way, each pixel
selecting unit 411-3 through pixel selecting unit 411-L supplies
the selected set of pixels to each estimated error calculating unit
412-3 through estimated error calculating unit 412-L.
The estimated error calculating unit 412-1 through estimated error
calculating unit 412-L detect the correlation of the pixels values
of the pixels at positions in the multiple sets, supplied from each
of the pixel selecting unit 411-1 through pixel selecting unit
411-L. The estimated error calculating unit 412-1 through estimated
error calculating unit 412-L supply information indicating the
detected correlation to the smallest error angle selecting unit
413.
The smallest error angle selecting unit 413 detects the data
continuity angle based on the reference axis in the input image
which corresponds to the continuity of the image which is the lost
actual world 1 light signals, based on the correlation detected by
the estimated error calculating unit 412-1 through estimated error
calculating unit 412-L.
Next, data continuity detection processing with the data continuity
detecting unit 101 of which the configuration is shown in FIG. 72,
corresponding to the processing in step S101, will be described
with reference to the flowchart in FIG. 79.
In step S401, the activity detecting unit 401 and the data
selecting unit 402 select the pixel of interest which is a pixel of
interest from the input image. The activity detecting unit 401 and
the data selecting unit 402 select the same pixel of interest. For
example, the activity detecting unit 401 and the data selecting
unit 402 select the pixel of interest from the input image in
raster scan order.
In step S402, the activity detecting unit 401 detects activity with
regard to the pixel of interest. For example, the activity
detecting unit 401 detects activity based on the difference of
pixel values of pixels aligned in the vertical direction of a block
made up of a predetermined number of pixels centered on the pixel
of interest, and the difference of pixel values of pixels aligned
in the horizontal direction.
The activity detecting unit 401 detects activity in the spatial
direction as to the pixel of interest, and supplies activity
information indicating the detected results to the data selecting
unit 402 and the continuity direction derivation unit 404.
In step S403, the data selecting unit 402 selects, from a row of
pixels including the pixel of interest, a predetermined number of
pixels centered on the pixel of interest, as a pixel set. For
example, the data selecting unit 402 selects a predetermined number
of pixels above or to the left of the pixel of interest, and a
predetermined number of pixels below or to the right of the pixel
of interest, which are pixels belonging to a vertical or horizontal
row of pixels to which the pixel of interest belongs, and also the
pixel of interest, as a pixel set.
In step S404, the data selecting unit 402 selects, as a pixel set,
a predetermined number of pixels each from a predetermined number
of pixel rows for each angle in a predetermined range based on the
activity detected by the processing in step S402. For example, the
data selecting unit 402 sets straight lines with angles of a
predetermined range which pass through the pixel of interest, with
the axis indicating the spatial direction X as the reference axis,
selects a pixel which is one or two rows away from the pixel of
interest in the horizontal direction or vertical direction and
which is closest to the straight line, and selects a predetermined
number of pixels above or to the left of the selected pixel, and a
predetermined number of pixels below or to the right of the
selected pixel, and the selected pixel closest to the line, as a
pixel set. The data selecting unit 402 selects pixel sets for each
angle.
The data selecting unit 402 supplies the selected pixel sets to the
error estimating unit 403.
In step S405, the error estimating unit 403 calculates the
correlation between the set of pixels centered on the pixel of
interest, and the pixel sets selected for each angle. For example,
the error estimating unit 403 calculates the sum of absolute values
of difference of the pixel values of the pixels of the set
including the pixel of interest and the pixel values of the pixels
at corresponding positions in other sets, for each angle.
The angle of data continuity may be detected based on the
correlation between pixel sets selected for each angle.
The error estimating unit 403 supplies the information indicating
the calculated correlation to the continuity direction derivation
unit 404.
In step S406, from position of the pixel set having the strongest
correlation based on the correlation calculated in the processing
in step S405, the continuity direction derivation unit 404 detects
the data continuity angle based on the reference axis in the input
image which is image data that corresponds to the lost actual world
1 light signal continuity. For example, the continuity direction
derivation unit 404 selects the smallest aggregate of the aggregate
of absolute values of difference of pixel values, and detects the
data continuity angle .theta. from the position of the pixel set
regarding which the selected aggregate has been calculated.
The continuity direction derivation unit 404 outputs data
continuity information indicating the angle of the data continuity
that has been detected.
In step S407, the data selecting unit 402 determines whether or not
processing of all pixels has ended, and in the event that
determination is made that processing of all pixels has not ended,
the flow returns to step S401, a pixel of interest is selected from
pixels not yet taken as the pixel of interest, and the
above-described processing is repeated.
In the event that determination is made in step S407 that
processing of all pixels has ended, the processing ends.
Thus, the data continuity detecting unit 101 can detect the data
continuity angle based on the reference axis in the image data,
corresponding to the lost actual world 1 light signal
continuity.
Note that an arrangement may be made wherein the data continuity
detecting unit 101 of which the configuration is shown in FIG. 72
detects activity in the spatial direction of the input image with
regard to the pixel of interest which is a pixel of interest in the
frame of interest which is a frame of interest, extracts multiple
pixel sets made up of a predetermined number of pixels in one row
in the vertical direction or one row in the horizontal direction
from the frame of interest and from each of frames before or after
time-wise the frame of interest, for each angle and movement vector
based on the pixel of interest and the space-directional reference
axis, according to the detected activity, detects the correlation
of the extracted pixel sets, and detects the data continuity angle
in the time direction and spatial direction in the input image,
based on this correlation.
For example, as shown in FIG. 80, the data selecting unit 402
extracts multiple pixel sets made up of a predetermined number of
pixels in one row in the vertical direction or one row in the
horizontal direction from frame #n which is the frame of interest,
frame #n-1, and frame #n+1, for each angle and movement vector
based on the pixel of interest and the space-directional reference
axis, according to the detected activity.
The frame #n-1 is a frame which is previous to the frame #n
time-wise, and the frame #n+1 is a frame following the frame #n
time-wise. That is to say, the frame #n-1, frame #n, and frame
#n+1, are displayed in the order of frame #n-1, frame #n, and frame
#n+1.
The error estimating unit 403 detects the correlation of pixel sets
for each single angle and single movement vector, with regard to
the multiple sets of the pixels that have been extracted. The
continuity direction derivation unit 404 detects the data
continuity angle in the temporal direction and spatial direction in
the input image which corresponds to the lost actual world 1 light
signal continuity, based on the correlation of pixel sets, and
outputs the data continuity information indicating the angle.
FIG. 81 is a block diagram illustrating another configuration of
the data continuity detecting unit 101 shown in FIG. 72, in further
detail. Portions which are the same as the case shown in FIG. 76
are denoted with the same numerals, and description thereof will be
omitted.
The data selecting unit 402 includes pixel selecting unit 421-1
through pixel selecting unit 421-L. The error estimating unit 403
includes estimated error calculating unit 422-1 through estimated
error calculating unit 422-L.
With the data continuity detecting unit 101 shown in FIG. 81, sets
of a number corresponding to the range of the angle are extracted
wherein the pixel sets are made up of pixels of a number
corresponding to the range of the angle, the correlation of the
extracted pixel sets is detected, and the data continuity angle
based on the reference axis in the input image is detected based on
the detected correlation.
First, the processing of the pixel selecting unit 421-1 through
pixel selecting unit 421-L in the event that the angle of the data
continuity indicated by activity information is any value 45
degrees to 135 degrees, will be described.
As shown to the left side in FIG. 82, with the data continuity
detecting unit 101 shown in FIG. 76, pixel sets of a predetermined
number of pixels are extracted regardless of the angle of the set
straight line, but with the data continuity detecting unit 101
shown in FIG. 81, pixel sets of a number of pixels corresponding to
the range of the angle of the set straight line are extracted, as
indicated at the right side of FIG. 82. Also, with the data
continuity detecting unit 101 shown in FIG. 81, pixels sets of a
number corresponding to the range of the angle of the set straight
line are extracted.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
set straight lines of mutually differing predetermined angles which
pass through the pixel of interest with the axis indicating the
spatial direction X as a reference axis, in the range of 45 degrees
to 135 degrees.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
select, from pixels belonging to one vertical row of pixels to
which the pixel of interest belongs, pixels above the pixel of
interest and pixels below the pixel of interest of a number
corresponding to the range of the angle of the straight line set
for each, and the pixel of interest, as a pixel set.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
select, from pixels belonging to one vertical line each on the left
side and the right side as to the one vertical row of pixels to
which the pixel of interest belongs, a predetermined distance away
therefrom in the horizontal direction with the pixel as a
reference, pixels closest to the straight lines set for each, and
selects, from one vertical row of pixels as to the selected pixel,
pixels above the selected pixel of a number corresponding to the
range of angle of the set straight line, pixels below the selected
pixel of a number corresponding to the range of angle of the set
straight line, and the selected pixel, as a pixel set.
That is to say, the pixel selecting unit 421-1 through pixel
selecting unit 421-L select pixels of a number corresponding to the
range of angle of the set straight line as pixel sets. The pixel
selecting unit 421-1 through pixel selecting unit 421-L select
pixels sets of a number corresponding to the range of angle of the
set straight line.
For example, in the event that the image of a fine line, positioned
at an angle approximately 45 degrees as to the spatial direction X,
and having a width which is approximately the same width as the
detection region of a detecting element, has been imaged with the
sensor 2, the image of the fine line is projected on the data 3
such that arc shapes are formed on three pixels aligned in one row
in the spatial direction Y for the fine-line image. Conversely, in
the event that the image of a fine line, positioned at an angle
approximately vertical to the spatial direction X, and having a
width which is approximately the same width as the detection region
of a detecting element, has been imaged with the sensor 2, the
image of the fine line is projected on the data 3 such that arc
shapes are formed on a great number of pixels aligned in one row in
the spatial direction Y for the fine-line image.
With the same number of pixels included in the pixel sets, in the
event that the fine line is positioned at an angle approximately 45
degrees to the spatial direction X, the number of pixels on which
the fine line image has been projected is smaller in the pixel set,
meaning that the resolution is lower. On the other hand, in the
event that the fine line is positioned approximately vertical to
the spatial direction X, processing is performed on a part of the
pixels on which the fine line image has been projected, which may
lead to lower accuracy.
Accordingly, to make the number of pixels upon which the fine line
image is projected to be approximately equal, the pixel selecting
unit 421-1 through pixel selecting unit 421-L selects the pixels
and the pixel sets so as to reduce the number of pixels included in
each of the pixels sets and increase the number of pixel sets in
the event that the straight line set is closer to an angle of 45
degrees as to the spatial direction X, and increase the number of
pixels included in each of the pixels sets and reduce the number of
pixel sets in the event that the straight line set is closer to
being vertical as to the spatial direction X.
For example, as shown in FIG. 83 and FIG. 84, in the event that the
angle of the set straight line is within the range of 45 degrees or
greater but smaller than 63.4 degrees (the range indicated by A in
FIG. 83 and FIG. 84), the pixel selecting unit 421-1 through pixel
selecting unit 421-L select five pixels centered on the pixel of
interest from one vertical row of pixels as to the pixel of
interest, as a pixel set, and also select as pixel sets five pixels
each from pixels belonging to one row of pixels each on the left
side and the right side of the pixel of interest within five pixels
therefrom in the horizontal direction.
That is to say, in the event that the angle of the set straight
line is within the range of 45 degrees or greater but smaller than
63.4 degrees the pixel selecting unit 421-1 through pixel selecting
unit 421-L select 11 pixel sets each made up of five pixels, from
the input image. In this case, the pixel selected as the pixel
which is at the closest position to the set straight line is at a
position five pixels to nine pixels in the vertical direction as to
the pixel of interest.
In FIG. 84, the number of rows indicates the number of rows of
pixels to the left side or right side of the pixel of interest from
which pixels are selected as pixel sets. In FIG. 84, the number of
pixels in one row indicates the number of pixels selected as a
pixel set from the one row of pixels vertical as to the pixel of
interest, or the rows to the left side or the right side of the
pixel of interest. In FIG. 84, the selection range of pixels
indicates the position of pixels to be selected in the vertical
direction, as the pixel at a position closest to the set straight
line as to the pixel of interest.
As shown in FIG. 85, for example, in the event that the angle of
the set straight line is 45 degrees, the pixel selecting unit 421-1
selects five pixels centered on the pixel of interest from one
vertical row of pixels as to the pixel of interest, as a pixel set,
and also selects as pixel sets five pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within five pixels therefrom in the
horizontal direction. That is to say, the pixel selecting unit
421-1 selects 11 pixel sets each made up of five pixels, from the
input image. In this case, of the pixels selected as the pixels at
the closest position to the set straight line the pixel which is at
the farthest position from the pixel of interest is at a position
five pixels in the vertical direction as to the pixel of
interest.
Note that in FIG. 85 through FIG. 92, the squares represented by
dotted lines (single grids separated by dotted lines) indicate
single pixels, and squares represented by solid lines indicate
pixel sets. In FIG. 85 through FIG. 92, the coordinate of the pixel
of interest in the spatial direction X is 0, and the coordinate of
the pixel of interest in the spatial direction Y is 0.
Also, in FIG. 85 through FIG. 92, the hatched squares indicate the
pixel of interest or the pixels at positions closest to the set
straight line. In FIG. 85 through FIG. 92, the squares represented
by heavy lines indicate the set of pixels selected with the pixel
of interest as the center.
As shown in FIG. 86, for example, in the event that the angle of
the set straight line is 60.9 degrees, the pixel selecting unit
421-2 selects five pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also selects as pixel sets five pixels each from pixels
belonging to one vertical row of pixels each on the left side and
the right side of the pixel of interest within five pixels
therefrom in the horizontal direction. That is to say, the pixel
selecting unit 421-2 selects 11 pixel sets each made up of five
pixels, from the input image. In this case, of the pixels selected
as the pixels at the closest position to the set straight line the
pixel which is at the farthest position from the pixel of interest
is at a position nine pixels in the vertical direction as to the
pixel of interest.
For example, as shown in FIG. 83 and FIG. 84, in the event that the
angle of the set straight line is 63.4 degrees or greater but
smaller than 71.6 degrees (the range indicated by B in FIG. 83 and
FIG. 84), the pixel selecting unit 421-1 through pixel selecting
unit 421-L select seven pixels centered on the pixel of interest
from one vertical row of pixels as to the pixel of interest, as a
pixel set, and also select as pixel sets seven pixels each from
pixels belonging to one row of pixels each on the left side and the
right side of the pixel of interest within four pixels therefrom in
the horizontal direction.
That is to say, in the event that the angle of the set straight
line is 63.4 degrees or greater but smaller than 71.6 degrees the
pixel selecting unit 421-1 through pixel selecting unit 421-L
select nine pixel sets each made up of seven pixels, from the input
image. In this case, the pixel selected as the pixel which is at
the closest position to the set straight line is at a position
eight pixels to 11 pixels in the vertical direction as to the pixel
of interest.
As shown in FIG. 87, for example, in the event that the angle of
the set straight line is 63.4 degrees, the pixel selecting unit
421-3 selects seven pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also selects as pixel sets seven pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within four pixels therefrom in the
horizontal direction. That is to say, the pixel selecting unit
421-3 selects nine pixel sets each made up of seven pixels, from
the input image. In this case, of the pixels selected as the pixels
at the closest position to the set straight line the pixel which is
at the farthest position from the pixel of interest is at a
position eight pixels in the vertical direction as to the pixel of
interest.
As shown in FIG. 88, for example, in the event that the angle of
the set straight line is 70.0 degrees, the pixel selecting unit
421-4 selects seven pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also selects as pixel sets seven pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within four pixels therefrom in the
horizontal direction. That is to say, the pixel selecting unit
421-4 selects nine pixel sets each made up of seven pixels, from
the input image. In this case, of the pixels selected as the pixels
at the closest position to the set straight line the pixel which is
at the farthest position from the pixel of interest is at a
position 11 pixels in the vertical direction as to the pixel of
interest.
For example, as shown in FIG. 83 and FIG. 84, in the event that the
angle of the set straight line is 71.6 degrees or greater but
smaller than 76.0 degrees (the range indicated by C in FIG. 83 and
FIG. 84), the pixel selecting unit 421-1 through pixel selecting
unit 421-L select nine pixels centered on the pixel of interest
from one vertical row of pixels as to the pixel of interest, as a
pixel set, and also select as pixel sets nine pixels each from
pixels belonging to one row of pixels each on the left side and the
right side of the pixel of interest within three pixels therefrom
in the horizontal direction.
That is to say, in the event that the angle of the set straight
line is 71.6 degrees or greater but smaller than 76.0 degrees, the
pixel selecting unit 421-1 through pixel selecting unit 421-L
select seven pixel sets each made up of nine pixels, from the input
image. In this case, the pixel selected as the pixel which is at
the closest position to the set straight line is at a position nine
pixels to 11 pixels in the vertical direction as to the pixel of
interest.
As shown in FIG. 89, for example, in the event that the angle of
the set straight line is 71.6 degrees, the pixel selecting unit
421-5 selects nine pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also selects as pixel sets nine pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within three pixels therefrom in the
horizontal direction. That is to say, the pixel selecting unit
421-5 selects seven pixel sets each made up of nine pixels, from
the input image. In this case, of the pixels selected as the pixels
at the closest position to the set straight line the pixel which is
at the farthest position from the pixel of interest is at a
position nine pixels in the vertical direction as to the pixel of
interest.
Also, As shown in FIG. 90, for example, in the event that the angle
of the set straight line is 74.7 degrees, the pixel selecting unit
421-6 selects nine pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also selects as pixel sets nine pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within three pixels therefrom in the
horizontal direction. That is to say, the pixel selecting unit
421-6 selects seven pixel sets each made up of nine pixels, from
the input image. In this case, of the pixels selected as the pixels
at the closest position to the set straight line the pixel which is
at the farthest position from the pixel of interest is at a
position 11 pixels in the vertical direction as to the pixel of
interest.
For example, as shown in FIG. 83 and FIG. 84, in the event that the
angle of the set straight line is 76.0 degrees or greater but
smaller than 87.7 degrees (the range indicated by D in FIG. 83 and
FIG. 84), the pixel selecting unit 421-1 through pixel selecting
unit 421-L select 11 pixels centered on the pixel of interest from
one vertical row of pixels as to the pixel of interest, as a pixel
set, and also select as pixel sets 11 pixels each from pixels
belonging to one row of pixels each on the left side and the right
side of the pixel of interest within two pixels therefrom in the
horizontal direction. That is to say, in the event that the angle
of the set straight line is 76.0 degrees or greater but smaller
than 87.7 degrees, the pixel selecting unit 421-1 through pixel
selecting unit 421-L select five pixel sets each made up of 11
pixels, from the input image. In this case, the pixel selected as
the pixel which is at the closest position to the set straight line
is at a position eight pixels to 50 pixels in the vertical
direction as to the pixel of interest.
As shown in FIG. 91, for example, in the event that the angle of
the set straight line is 76.0 degrees, the pixel selecting unit
421-7 selects 11 pixels centered on the pixel of interest from one
vertical row of pixels as to the pixel of interest, as a pixel set,
and also selects as pixel sets 11 pixels each from pixels belonging
to one row of pixels each on the left side and the right side of
the pixel of interest within two pixels therefrom in the horizontal
direction. That is to say, the pixel selecting unit 421-7 selects
five pixel sets each made up of 11 pixels, from the input image. In
this case, of the pixels selected as the pixels at the closest
position to the set straight line the pixel which is at the
farthest position from the pixel of interest is at a position eight
pixels in the vertical direction as to the pixel of interest.
Also, as shown in FIG. 92, for example, in the event that the angle
of the set straight line is 87.7 degrees, the pixel selecting unit
421-8 selects 11 pixels centered on the pixel of interest from one
vertical row of pixels as to the pixel of interest, as a pixel set,
and also selects as pixel sets 11 pixels each from pixels belonging
to one row of pixels each on the left side and the right side of
the pixel of interest within two pixels therefrom in the horizontal
direction. That is to say, the pixel selecting unit 421-8 selects
five pixel sets each made up of 11 pixels, from the input image. In
this case, of the pixels selected as the pixels at the closest
position to the set straight line the pixel which is at the
farthest position from the pixel of interest is at a position 50
pixels in the vertical direction as to the pixel of interest.
Thus, the pixel selecting unit 421-1 through pixel selecting unit
421-L each select a predetermined number of pixels sets
corresponding to the range of the angle, made up of a predetermined
number of pixels corresponding to the range of the angle.
The pixel selecting unit 421-1 supplies the selected pixel sets to
an estimated error calculating unit 422-1, and the pixel selecting
unit 421-2 supplies the selected pixel sets to an estimated error
calculating unit 422-2. In the same way, the pixel selecting unit
421-3 through pixel selecting unit 421-L supply the selected pixel
sets to estimated error calculating unit 422-3 through estimated
error calculating unit 422-L.
The estimated error calculating unit 422-1 through estimated error
calculating unit 422-L detect the correlation of pixel values of
the pixels at corresponding positions in the multiple sets supplied
from each of the pixel selecting unit 421-1 through pixel selecting
unit 421-L. For example, the estimated error calculating unit 422-1
through estimated error calculating unit 422-L calculate the sum of
absolute values of difference between the pixel values of the
pixels of the pixel set including the pixel of interest, and of the
pixel values of the pixels at corresponding positions in the other
multiple sets, supplied from each of the pixel selecting unit 421-1
through pixel selecting unit 421-L, and divides the calculated sum
by the number of pixels contained in the pixel sets other than the
pixel set containing the pixel of interest. The reason for dividing
the calculated sum by the number of pixels contained in sets other
than the set containing the pixel of interest is to normalize the
value indicating the correlation, since the number of pixels
selected differs according to the angle of the straight line that
has been set.
The estimated error calculating unit 422-1 through estimated error
calculating unit 422-L supply the detected information indicating
correlation to the smallest error angle selecting unit 413. For
example, the estimated error calculating unit 422-1 through
estimated error calculating unit 422-L supply the normalized sum of
difference of the pixel values to the smallest error angle
selecting unit 413.
Next, the processing of the pixel selecting unit 421-1 through
pixel selecting unit 421-L in the event that the angle of the data
continuity indicated by activity information is any value 0 degrees
to 45 degrees and 135 degrees to 180 degrees, will be
described.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
set straight lines of mutually differing predetermined angles which
pass through the pixel of interest with the axis indicating the
spatial direction X as a reference, in the range of 0 degrees to 45
degrees or 135 degrees to 180 degrees.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
select, from pixels belonging to one horizontal row of pixels to
which the pixel of interest belongs, pixels to the left side of the
pixel of interest of a number corresponding to the range of angle
of the set line, pixels to the right side of the pixel of interest
of a number corresponding to the range of angle of the set line,
and the selected pixel, as a pixel set.
The pixel selecting unit 421-1 through pixel selecting unit 421-L
select, from pixels belonging to one horizontal line each above and
below as to the one horizontal row of pixels to which the pixel of
interest belongs, a predetermined distance away therefrom in the
vertical direction with the pixel as a reference, pixels closest to
the straight lines set for each, and selects, from one horizontal
row of pixels as to the selected pixel, pixels to the left side of
the selected pixel of a number corresponding to the range of angle
of the set line, pixels to the right side of the selected pixel of
a number corresponding to the range of angle of the set line, and
the selected pixel, as a pixel set.
That is to say, the pixel selecting unit 421-1 through pixel
selecting unit 421-L select pixels of a number corresponding to the
range of angle of the set line as pixel sets. The pixel selecting
unit 421-1 through pixel selecting unit 421-L select pixels sets of
a number corresponding to the range of angle of the set line.
The pixel selecting unit 421-1 supplies the selected set of pixels
to the estimated error calculating unit 422-1, and the pixel
selecting unit 421-2 supplies the selected set of pixels to the
estimated error calculating unit 422-2. In the same way, each pixel
selecting unit 421-3 through pixel selecting unit 421-L supplies
the selected set of pixels to each estimated error calculating unit
422-3 through estimated error calculating unit 422-L.
The estimated error calculating unit 422-1 through estimated error
calculating unit 422-L detect the correlation of pixel values of
the pixels at corresponding positions in the multiple sets supplied
from each of the pixel selecting unit 421-1 through pixel selecting
unit 421-L.
The estimated error calculating unit 422-1 through estimated error
calculating unit 422-L supply the detected information indicating
correlation to the smallest error angle selecting unit 413.
Next, the processing for data continuity detection with the data
continuity detecting unit 101 of which the configuration is shown
in FIG. 81, corresponding to the processing in step S101, will be
described with reference to the flowchart shown in FIG. 93.
The processing of step S421 and step S422 is the same as the
processing of step S401 and step S402, so description thereof will
be omitted.
In step S423, the data selecting unit 402 selects, from a row of
pixels containing a pixel of interest, a number of pixels
predetermined with regard to the range of the angle which are
centered on the pixel of interest, as a set of pixels, for each
angle of a range corresponding to the activity detected in the
processing in step S422. For example, the data selecting unit 402
selects from pixels belonging to one vertical or horizontal row of
pixels, pixels of a number determined by the range of angle, for
the angle of the straight line to be set, above or to the left of
the pixel of interest, below or to the right of the pixel of
interest, and the pixel of interest, as a pixel set.
In step S424, the data selecting unit 402 selects, from pixel rows
of a number determined according to the range of angle, pixels of a
number determined according to the range of angle, as a pixel set,
for each predetermined angle range, based on the activity detected
in the processing in step S422. For example, the data selecting
unit 402 sets a straight line passing through the pixel of interest
with an angle of a predetermined range, taking an axis representing
the spatial direction X as a reference axis, selects a pixel
closest to the straight line while being distanced from the pixel
of interest in the horizontal direction or the vertical direction
by a predetermined range according to the range of angle of the
straight line to be set, and selects pixels of a number
corresponding to the range of angle of the straight line to be set
from above or to the left side of the selected pixel, pixels of a
number corresponding to the range of angle of the straight line to
be set from below or to the right side of the selected pixel, and
the pixel closest to the selected line, as a pixel set. The data
selecting unit 402 selects a set of pixels for each angle.
The data selecting unit 402 supplies the selected pixel sets to the
error estimating unit 403.
In step S425, the error estimating unit 403 calculates the
correlation between the pixel set centered on the pixel of
interest, and the pixel set selected for each angle. For example,
the error estimating unit 403 calculates the sum of absolute values
of difference between the pixel values of pixels of the set
including the pixel of interest and the pixel values of pixels at
corresponding positions in the other sets, and divides the sum of
absolute values of difference between the pixel values by the
number of pixels belonging to the other sets, thereby calculating
the correlation.
An arrangement may be made wherein the data continuity angle is
detected based on the mutual correlation between the pixel sets
selected for each angle.
The error estimating unit 403 supplies the information indicating
the calculated correlation to the continuity direction derivation
unit 404.
The processing of step S426 and step S427 is the same as the
processing of step S406 and step S407, so description thereof will
be omitted.
Thus, the data continuity detecting unit 101 can detect the angle
of data continuity based on a reference axis in the image data,
corresponding to the lost actual world 1 light signal continuity,
more accurately and precisely. With the data continuity detecting
unit 101 of which the configuration is shown in FIG. 81, the
correlation of a greater number of pixels where the fine line image
has been projected can be evaluated particularly in the event that
the data continuity angle is around 45 degrees, so the angle of
data continuity can be detected with higher precision.
Note that an arrangement may be made with the data continuity
detecting unit 101 of which the configuration is shown in FIG. 81
as well, wherein activity in the spatial direction of the input
image is detected for a certain pixel of interest which is the
pixel of interest in a frame of interest which is the frame of
interest, and from sets of pixels of a number determined according
to the spatial angle range in one vertical row or one horizontal
row, pixels of a number corresponding to the spatial angle range
are extracted, from the frame of interest and frames previous to or
following the frame of interest time-wise, for each angle and
movement vector based on the pixel of interest and the reference
axis in the spatial direction, according to the detected activity,
the correlation of the extracted pixel sets is detected, and the
data continuity angle in the time direction and the spatial
direction in the input image is detected based on the
correlation.
FIG. 94 is a block diagram illustrating yet another configuration
of the data continuity detecting unit 101.
With the data continuity detecting unit 101 of which the
configuration is shown in FIG. 94, with regard to a pixel of
interest which is the pixel of interest, a block made up of a
predetermined number of pixels centered on the pixel of interest,
and multiple blocks each made up of a predetermined number of
pixels around the pixel of interest, are extracted, the correlation
of the block centered on the pixel of interest and the surrounding
blocks is detected, and the angle of data continuity in the input
image based on a reference axis is detected, based on the
correlation.
A data selecting unit 441 sequentially selects the pixel of
interest from the pixels of the input image, extracts the block
made of the predetermined number of pixels centered on the pixel of
interest and the multiple blocks made up of the predetermined
number of pixels surrounding the pixel of interest, and supplies
the extracted blocks to an error estimating unit 442.
For example, the data selecting unit 441 extracts a block made up
of 5.times.5 pixels centered on the pixel of interest, and two
blocks made up of 5.times.5 pixels from the surroundings of the
pixel of interest for each predetermined angle range based on the
pixel of interest and the reference axis.
The error estimating unit 442 detects the correlation between the
block centered on the pixel of interest and the blocks in the
surroundings of the pixel of the interest supplied from the data
selecting unit 441, and supplies correlation information indicating
the detected correlation to a continuity direction derivation unit
443.
For example, the error estimating unit 442 detects the correlation
of pixel values with regard to a block made up of 5.times.5 pixels
centered on the pixel of interest for each angle range, and two
blocks made up of 5.times.5 pixels corresponding to one angle
range.
From the position of the block in the surroundings of the pixel of
interest with the greatest correlation based on the correlation
information supplied from the error estimating unit 442, the
continuity direction derivation unit 443 detects the angle of data
continuity in the input image based on the reference axis, that
corresponds to the lost actual world 1 light signal continuity, and
outputs data continuity information indicating this angle. For
example, the continuity direction derivation unit 443 detects the
range of the angle regarding the two blocks made up of 5.times.5
pixels from the surroundings of the pixel of interest which have
the greatest correlation with the block made up of 5.times.5 pixels
centered on the pixel of interest, as the angle of data continuity,
based on the correlation information supplied from the error
estimating unit 442, and outputs data continuity information
indicating the detected angle.
FIG. 95 is a block diagram illustrating a more detailed
configuration of the data continuity detecting unit 101 shown in
FIG. 94.
The data selecting unit 441 includes pixel selecting unit 461-1
through pixel selecting unit 461-L. The error estimating unit 442
includes estimated error calculating unit 462-1 through estimated
error calculating unit 462-L. The continuity direction derivation
unit 443 includes a smallest error angle selecting unit 463.
For example, the data selecting unit 441 has pixel selecting unit
461-1 through pixel selecting unit 461-8. The error estimating unit
442 has estimated error calculating unit 462-1 through estimated
error calculating unit 462-8.
Each of the pixel selecting unit 461-1 through pixel selecting unit
461-L extracts a block made up of a predetermined number of pixels
centered on the pixel of interest, and two blocks made up of a
predetermined number of pixels according to a predetermined angle
range based on the pixel of interest and the reference axis.
FIG. 96 is a diagram for describing an example of a 5.times.5 pixel
block extracted by the pixel selecting unit 461-1 through pixel
selecting unit 461-L. The center position in FIG. 96 indicates the
position of the pixel of interest.
Note that a 5.times.5 pixel block is only an example, and the
number of pixels contained in a block do not restrict the present
invention.
For example, the pixel selecting unit 461-1 extracts a 5.times.5
pixel block centered on the pixel of interest, and also extracts a
5.times.5 pixel block (indicated by A in FIG. 96) centered on a
pixel at a position shifted five pixels to the right side from the
pixel of interest, and extracts a 5.times.5 pixel block (indicated
by A' in FIG. 96) centered on a pixel at a position shifted five
pixels to the left side from the pixel of interest, corresponding
to 0 degrees to 18.4 degrees and 161.6 degrees to 180.0 degrees.
The pixel selecting unit 461-1 supplies the three extracted
5.times.5 pixel blocks to the estimated error calculating unit
462-1.
The pixel selecting unit 461-2 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by B in FIG. 96) centered on a pixel at a
position shifted 10 pixels to the right side from the pixel of
interest and five pixels upwards, and extracts a 5.times.5 pixel
block (indicated by B' in FIG. 96) centered on a pixel at a
position shifted 10 pixels to the left side from the pixel of
interest and five pixels downwards, corresponding to the range of
18.4 degrees through 33.7 degrees. The pixel selecting unit 461-2
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-2.
The pixel selecting unit 461-3 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by C in FIG. 96) centered on a pixel at a
position shifted five pixels to the right side from the pixel of
interest and five pixels upwards, and extracts a 5.times.5 pixel
block (indicated by C' in FIG. 96) centered on a pixel at a
position shifted five pixels to the left side from the pixel of
interest and five pixels downwards, corresponding to the range of
33.7 degrees through 56.3 degrees. The pixel selecting unit 461-3
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-3.
The pixel selecting unit 461-4 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by D in FIG. 96) centered on a pixel at a
position shifted five pixels to the right side from the pixel of
interest and 10 pixels upwards, and extracts a 5.times.5 pixel
block (indicated by D' in FIG. 96) centered on a pixel at a
position shifted five pixels to the left side from the pixel of
interest and 10 pixels downwards, corresponding to the range of
56.3 degrees through 71.6 degrees. The pixel selecting unit 461-4
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-4.
The pixel selecting unit 461-5 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by E in FIG. 96) centered on a pixel at a
position shifted five pixels upwards from the pixel of interest,
and extracts a 5.times.5 pixel block (indicated by E' in FIG. 96)
centered on a pixel at a position shifted five pixels downwards
from the pixel of interest, corresponding to the range of 71.6
degrees through 108.4 degrees. The pixel selecting unit 461-5
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-5.
The pixel selecting unit 461-6 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by F in FIG. 96) centered on a pixel at a
position shifted five pixels to the left side from the pixel of
interest and 10 pixels upwards, and extracts a 5.times.5 pixel
block (indicated by F' in FIG. 96) centered on a pixel at a
position shifted five pixels to the right side from the pixel of
interest and 10 pixels downwards, corresponding to the range of
108.4 degrees through 123.7 degrees. The pixel selecting unit 461-6
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-6.
The pixel selecting unit 461-7 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by G in FIG. 96) centered on a pixel at a
position shifted five pixels to the left side from the pixel of
interest and five pixels upwards, and extracts a 5.times.5 pixel
block (indicated by G' in FIG. 96) centered on a pixel at a
position shifted five pixels to the right side from the pixel of
interest and five pixels downwards, corresponding to the range of
123.7 degrees through 146.3 degrees. The pixel selecting unit 461-7
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-7.
The pixel selecting unit 461-8 extracts a 5.times.5 pixel block
centered on the pixel of interest, and also extracts a 5.times.5
pixel block (indicated by H in FIG. 96) centered on a pixel at a
position shifted 10 pixels to the left side from the pixel of
interest and five pixels upwards, and extracts a 5.times.5 pixel
block (indicated by H' in FIG. 96) centered on a pixel at a
position shifted 10 pixels to the right side from the pixel of
interest and five pixels downwards, corresponding to the range of
146.3 degrees through 161.6 degrees. The pixel selecting unit 461-8
supplies the three extracted 5.times.5 pixel blocks to the
estimated error calculating unit 462-8.
Hereafter, a block made up of a predetermined number of pixels
centered on the pixel of interest will be called a block of
interest.
Hereafter, a block made up of a predetermined number of pixels
corresponding to a predetermined range of angle based on the pixel
of interest and reference axis will be called a reference
block.
In this way, the pixel selecting unit 461-1 through pixel selecting
unit 461-8 extract a block of interest and reference blocks from a
range of 25.times.25 pixels, centered on the pixel of interest, for
example.
The estimated error calculating unit 462-1 through estimated error
calculating unit 462-L detect the correlation between the block of
interest and the two reference blocks supplied from the pixel
selecting unit 461-1 through pixel selecting unit 461-L, and
supplies correlation information indicating the detected
correlation to the smallest error angle selecting unit 463.
For example, the estimated error calculating unit 462-1 calculates
the absolute value of difference between the pixel values of the
pixels contained in the block of interest and the pixel values of
the pixels contained in the reference block, with regard to the
block of interest made up of 5.times.5 pixels centered on the pixel
of interest, and the 5.times.5 pixel reference block centered on a
pixel at a position shifted five pixels to the right side from the
pixel of interest, extracted corresponding to 0 degrees to 18.4
degrees and 161.6 degrees to 180.0 degrees.
In this case, as shown in FIG. 97, in order for the pixel value of
the pixel of interest to be used on the calculation of the absolute
value of difference of pixel values, with the position where the
center pixel of the block of interest and the center pixel of the
reference block overlap as a reference, the estimated error
calculating unit 462-1 calculates the absolute value of difference
of pixel values of pixels at positions overlapping in the event
that the position of the block of interest is shifted to any one of
two pixels to the left side through two pixels to the right side
and any one of two pixels upwards through two pixels downwards as
to the reference block. This means that the absolute value of
difference of the pixel values of pixels at corresponding positions
in 25 types of positions of the block of interest and the reference
block. In other words, in a case wherein the absolute values of
difference of the pixel values are calculated, the range formed of
the block of interest moved relatively and the reference block is
9.times.9 pixels.
In FIG. 97, the square represent pixels, A represents the reference
block, and B represents the block of interest. In FIG. 97, the
heavy lines indicate the pixel of interest. That is to say, FIG. 97
is a diagram illustrating a case wherein the block of interest has
been shifted two pixels to the right side and one pixel upwards, as
to the reference block.
Further, the estimated error calculating unit 462-1 calculates the
absolute value of difference between the pixel values of the pixels
contained in the block of interest and the pixel values of the
pixels contained in the reference block, with regard to the block
of interest made up of 5.times.5 pixels centered on the pixel of
interest, and the 5.times.5 pixel reference block centered on a
pixel at a position shifted five pixels to the left side from the
pixel of interest, extracted corresponding to 0 degrees to 18.4
degrees and 161.6 degrees to 180.0 degrees.
The estimated error calculating unit 462-1 then obtains the sum of
the absolute values of difference that have been calculated, and
supplies the sum of the absolute values of difference to the
smallest error angle selecting unit 463 as correlation information
indicating correlation.
The estimated error calculating unit 462-2 calculates the absolute
value of difference between the pixel values with regard to the
block of interest made up of 5.times.5 pixels and the two 5.times.5
reference pixel blocks extracted corresponding to the range of 18.4
degrees to 33.7 degrees, and further calculates sum of the absolute
values of difference that have been calculated. The estimated error
calculating unit 462-1 supplies the sum of the absolute values of
difference that has been calculated to the smallest error angle
selecting unit 463 as correlation information indicating
correlation.
In the same way, the estimated error calculating unit 462-3 through
estimated error calculating unit 462-8 calculate the absolute value
of difference between the pixel values with regard to the block of
interest made up of 5.times.5 pixels and the two 5.times.5 pixel
reference blocks extracted corresponding to the predetermined angle
ranges, and further calculate sum of the absolute values of
difference that have been calculated. The estimated error
calculating unit 462-3 through estimated error calculating unit
462-8 each supply the sum of the absolute values of difference to
the smallest error angle selecting unit 463 as correlation
information indicating correlation.
The smallest error angle selecting unit 463 detects, as the data
continuity angle, the angle corresponding to the two reference
blocks at the reference block position where, of the sums of the
absolute values of difference of pixel values serving as
correlation information supplied from the estimated error
calculating unit 462-1 through estimated error calculating unit
462-8, the smallest value indicating the strongest correlation has
been obtained, and outputs data continuity information indicating
the detected angle.
Now, description will be made regarding the relationship between
the position of the reference blocks and the range of angle of data
continuity.
In a case of approximating an approximation function f(x) for
approximating actual world signals with an n-order one-dimensional
polynomial, the approximation function f(x) can be expressed by
Expression (30).
.function..times..times..times..times..times. ##EQU00017##
In the event that the waveform of the signal of the actual world 1
approximated by the approximation function f(x) has a certain
gradient (angle) as to the spatial direction Y, the approximation
function (x, y) for approximating actual world 1 signals is
expressed by Expression (31) which has been obtained by taking x in
Expression (30) as x+.gamma.y.
.function..function..gamma..times..times..function..gamma..times..times..-
function..gamma..times..times..times..function..gamma..times..times.
##EQU00018##
.gamma. represents the ratio of change in position in the spatial
direction X as to the change in position in the spatial direction
Y. Hereafter, .gamma. will also be called amount of shift.
FIG. 98 is a diagram illustrating the distance to a straight line
having an angle .theta. in the spatial direction X from the
position of surrounding pixels of the pixel of interest in a case
wherein the distance in the spatial direction X between the
position of the pixel of interest and the straight line having the
angle .theta. is 0, i.e., wherein the straight line passes through
the pixel of interest. Here, the position of the pixel is the
center of the pixel. Also, in the event that the position is to the
left side of the straight line, the distance between the position
and the straight line is indicated by a negative value, and in the
event that the position is to the right side of the straight line,
is indicated by a positive value.
For example, the distance in the spatial direction X between the
position of the pixel adjacent to the pixel of interest on the
right side, i.e., the position where the coordinate x in the
spatial direction X increases by 1, and the straight line having
the angle .theta., is 1, and the distance in the spatial direction
X between the position of the pixel adjacent to the pixel of
interest on the left side, i.e., the position where the coordinate
x in the spatial direction X decreases by 1, and the straight line
having the angle .theta., is -1. The distance in the spatial
direction X between the position of the pixel adjacent to the pixel
of interest above, i.e., the position where the coordinate y in the
spatial direction Y increases by 1, and the straight line having
the angle .theta., is -.gamma., and the distance in the spatial
direction X between the position of the pixel adjacent to the pixel
of interest below, i.e., the position where the coordinate y in the
spatial direction Y decreases by 1, and the straight line having
the angle .theta., is .gamma..
In the event that the angle .theta. exceeds 45 degrees but is
smaller than 90 degrees, and the amount of shift .gamma. exceeds 0
but is smaller than 1, the relational expression of .gamma.=1/tan
.theta. holds between the amount of shift .gamma. and the angle
.theta.. FIG. 99 is a diagram illustrating the relationship between
the amount of shift .gamma. and the angle .theta..
Now, let us take note of the change in distance in the spatial
direction X between the position of a pixel nearby the pixel of
interest, and the straight line which passes through the pixel of
interest and has the angle .theta., as to change in the amount of
shift .gamma..
FIG. 100 is a diagram illustrating the distance in the spatial
direction X between the position of a pixel nearby the pixel of
interest and the straight line which passes through the pixel of
interest and has the angle .theta., as to the amount of shift
.gamma.. In FIG. 100, the single-dot broken line which heads toward
the upper right indicates the distance in the spatial direction X
between the position of a pixel adjacent to the pixel of interest
on the bottom side, and the straight line, as to the amount of
shift .gamma.. The single-dot broken line which heads toward the
lower left indicates the distance in the spatial direction X
between the position of a pixel adjacent to the pixel of interest
on the top side, and the straight line, as to the amount of shift
.gamma..
In FIG. 100, the two-dot broken line which heads toward the upper
right indicates the distance in the spatial direction X between the
position of a pixel two pixels below the pixel of interest and one
to the left, and the straight line, as to the amount of shift
.gamma.; the two-dot broken line which heads toward the lower left
indicates the distance in the spatial direction X between the
position of a pixel two pixels above the pixel of interest and one
to the right, and the straight line, as to the amount of shift
.gamma..
In FIG. 100, the three-dot broken line which heads toward the upper
right indicates the distance in the spatial direction X between the
position of a pixel one pixel below the pixel of interest and one
to the left, and the straight line, as to the amount of shift
.gamma.; the three-dot broken line which heads toward the lower
left indicates the distance in the spatial direction X between the
position of a pixel one pixel above the pixel of interest and one
to the right, and the straight line, as to the amount of shift
y.
The pixel with the smallest distance as to the amount of shift
.gamma. can be found from FIG. 100.
That is to say, in the event that the amount of shift y is 0
through 1/3, the distance to the straight line is minimal from a
pixel adjacent to the pixel of interest on the top side and from a
pixel adjacent to the pixel of interest on the bottom side. That is
to say, in the event that the angle .theta. is 71.6 degrees to 90
degrees, the distance to the straight line is minimal from the
pixel adjacent to the pixel of interest on the top side and from
the pixel adjacent to the pixel of interest on the bottom side.
In the event that the amount of shift .gamma. is 1/3 through 2/3,
the distance to the straight line is minimal from a pixel two
pixels above the pixel of interest and one to the right and from a
pixel two pixels below the pixel of interest and one to the left.
That is to say, in the event that the angle .theta. is 56.3 degrees
to 71.6 degrees, the distance to the straight line is minimal from
the pixel two pixels above the pixel of interest and one to the
right and from a pixel two pixels below the pixel of interest and
one to the left.
In the event that the amount of shift .gamma. is 2/3 through 1, the
distance to the straight line is minimal from a pixel one pixel
above the pixel of interest and one to the right and from a pixel
one pixel below the pixel of interest and one to the left. That is
to say, in the event that the angle .theta. is 45 degrees to 56.3
degrees, the distance to the straight line is minimal from the
pixel one pixel above the pixel of interest and one to the right
and from a pixel one pixel below the pixel of interest and one to
the left.
The relationship between the straight line in a range of angle
.theta. from 0 degrees to 45 degrees and a pixel can also be
considered in the same way.
The pixels shown in FIG. 98 can be replaced with the block of
interest and reference block, to consider the distance in the
spatial direction X between the reference block and the straight
line.
FIG. 101 shows the reference blocks wherein the distance to the
straight line which passes through the pixel of interest and has an
angle .theta. as to the axis of the spatial direction X is the
smallest.
A through H and A' through H' in FIG. 101 represent the reference
blocks A through H and A' through H' in FIG. 96.
That is to say, of the distances in the spatial direction X between
a straight line having an angle .theta. which is any of 0 degrees
through 18.4 degrees and 161.6 degrees through 180.0 degrees which
passes through the pixel of interest with the axis of the spatial
direction X as a reference, and each of the reference blocks A
through H and A' through H', the distance between the straight line
and the reference blocks A and A' is the smallest. Accordingly,
following reverse logic, in the event that the correlation between
the block of interest and the reference blocks A and A' is the
greatest, this means that a certain feature is repeatedly
manifested in the direction connecting the block of interest and
the reference blocks A and A', so it can be said that the angle of
data continuity is within the ranges of 0 degrees through 18.4
degrees and 161.6 degrees through 180.0 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 18.4 degrees through 33.7
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks B and B' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
B and B' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks B and B', so it can be said that
the angle of data continuity is within the range of 18.4 degrees
through 33.7 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 33.7 degrees through 56.3
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks C and C' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
C and C' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks C and C', so it can be said that
the angle of data continuity is within the range of 33.7 degrees
through 56.3 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 56.3 degrees through 71.6
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks D and D' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
D and D' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks D and D', so it can be said that
the angle of data continuity is within the range of 56.3 degrees
through 71.6 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 71.6 degrees through 108.4
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks E and E' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
E and E' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks E and E', so it can be said that
the angle of data continuity is within the range of 71.6 degrees
through 108.4 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 108.4 degrees through 123.7
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks F and F' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
F and F' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks F and F', so it can be said that
the angle of data continuity is within the range of 108.4 degrees
through 123.7 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 123.7 degrees through 146.3
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks G and G' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
G and G' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks G and G', so it can be said that
the angle of data continuity is within the range of 123.7 degrees
through 146.3 degrees.
Of the distances in the spatial direction X between a straight line
having an angle .theta. which is any of 146.3 degrees through 161.6
degrees which passes through the pixel of interest with the axis of
the spatial direction X as a reference, and each of the reference
blocks A through H and A' through H', the distance between the
straight line and the reference blocks H and H' is the smallest.
Accordingly, following reverse logic, in the event that the
correlation between the block of interest and the reference blocks
H and H' is the greatest, this means that a certain feature is
repeatedly manifested in the direction connecting the block of
interest and the reference blocks H and H', so it can be said that
the angle of data continuity is within the range of 146.3 degrees
through 161.6 degrees.
Thus, the data continuity detecting unit 101 can detect the data
continuity angle based on the correlation between the block of
interest and the reference blocks.
Note that with the data continuity detecting unit 101 of which the
configuration is shown in FIG. 94, an arrangement may be made
wherein the angle range of data continuity is output as data
continuity information, or an arrangement may be made wherein a
representative value representing the range of angle of the data
continuity is output as data continuity information. For example,
the median value of the range of angle of the data continuity may
serve as a representative value.
Further, with the data continuity detecting unit 101 of which the
configuration is shown in FIG. 94, using the correlation between
the block of interest and the reference blocks with the greatest
correlation allows the angle range of data continuity to be
detected to be halved, i.e., for the resolution of the angle of
data continuity to be detected to be doubled.
For example, when the correlation between the block of interest and
the reference blocks E and E' is the greatest, the smallest error
angle selecting unit 463 compares the correlation of the reference
blocks D and D' as to the block of interest with the correlation of
the reference blocks F and F' as to the block of interest, as shown
in FIG. 102. In the event that the correlation of the reference
blocks D and D' as to the block of interest is greater than the
correlation of the reference blocks F and F' as to the block of
interest, the smallest error angle selecting unit 463 sets the
range of 71.6 degrees to 90 degrees for the data continuity angle.
Or, in this case, the smallest error angle selecting unit 463 may
set 81 degrees for the data continuity angle as a representative
value.
In the event that the correlation of the reference blocks F and F'
as to the block of interest is greater than the correlation of the
reference blocks D and D' as to the block of interest, the smallest
error angle selecting unit 463 sets the range of 90 degrees to
108.4 degrees for the data continuity angle. Or, in this case, the
smallest error angle selecting unit 463 may set 99 degrees for the
data continuity angle as a representative value.
The smallest error angle selecting unit 463 can halve the range of
the data continuity angle to be detected for other angle ranges as
well, with the same processing.
The technique described with reference to FIG. 102 is also called
simplified 16-directional detection.
Thus, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 94 can detect the angle of data
continuity in narrower ranges, with simple processing.
Next, the processing for detecting data continuity with the data
continuity detecting unit 101 of which the configuration is shown
in FIG. 94, corresponding to the processing in step S101, will be
described with reference to the flowchart shown in FIG. 103.
In step S441, the data selecting unit 441 selects the pixel of
interest which is a pixel of interest from the input image. For
example, the data selecting unit 441 selects the pixel of interest
in raster scan order from the input image.
In step S442, the data selecting unit 441 selects a block of
interest made up of a predetermined number of pixels centered on
the pixel of interest. For example, the data selecting unit 441
selects a block of interest made up of 5.times.5 pixels centered on
the pixel of interest.
In step S443, the data selecting unit 441 selects reference blocks
made up of a predetermined number of pixels at predetermined
positions at the surroundings of the pixel of interest. For
example, the data selecting unit 441 selects reference blocks made
up of 5.times.5 pixels centered on pixels at predetermined
positions based on the size of the block of interest, for each
predetermined angle range based on the pixel of interest and the
reference axis.
The data selecting unit 441 supplies the block of interest and the
reference blocks to the error estimating unit 442.
In step S444, the error estimating unit 442 calculates the
correlation between the block of interest and the reference blocks
corresponding to the range of angle, for each predetermined angle
range based on the pixel of interest and the reference axis. The
error estimating unit 442 supplies the correlation information
indicating the calculated correlation to the continuity direction
derivation unit 443.
In step S445, the continuity direction derivation unit 443 detects
the angle of data continuity in the input image based on the
reference axis, corresponding to the image continuity which is the
lost actual world 1 light signals, from the position of the
reference block which has the greatest correlation as to the block
of interest.
The continuity direction derivation unit 443 outputs the data
continuity information which indicates the detected data continuity
angle.
In step S446, the data selecting unit 441 determines whether or not
processing of all pixels has ended, and in the event that
determination is made that processing of all pixels has not ended,
the flow returns to step S441, a pixel of interest is selected from
pixels not yet selected as the pixel of interest, and the
above-described processing is repeated.
In step S446, in the event that determination is made that
processing of all pixels has ended, the processing ends.
Thus, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 94 can detect the data continuity
angle in the image data based on the reference axis, corresponding
to the lost actual world 1 light signal continuity with easier
processing. Also, the data continuity detecting unit 101 of which
the configuration is shown in FIG. 94 can detect the angle of data
continuity using pixel values of pixels of a relatively narrow
range in the input image, so the angle of data continuity can be
detected more accurately even in the event that noise and the like
is in the input image.
Note that an arrangement may be made with the data continuity
detecting unit 101 of which the configuration is shown in FIG. 94,
wherein, with regard to a pixel of interest which is the pixel of
interest in a frame of interest which is the frame of interest, in
addition to extracting a block centered on the pixel of interest
and made up of a predetermined number of pixels in the frame of
interest, and multiple blocks each made up of a predetermined
number of pixels from the surroundings of the pixel of interest,
also extracting, from frames previous to or following the frame of
interest time-wise, a block centered on a pixel at a position
corresponding to the pixel of interest and made up of a
predetermined number of pixels, and multiple blocks each made up of
a predetermined number of pixels from the surroundings of the pixel
centered on the pixel corresponding to the pixel of interest, and
detecting the correlation between the block centered on the pixel
of interest and blocks in the surroundings thereof space-wise or
time-wise, so as to detect the angle of data continuity in the
input image in the temporal direction and spatial direction, based
on the correlation.
For example, as shown in FIG. 104, the data selecting unit 441
sequentially selects the pixel of interest from the frame #n which
is the frame of interest, and extracts from the frame #n a block
centered on the pixel of interest and made up of a predetermined
number of pixels and multiple blocks each made up of a
predetermined number of pixels from the surroundings of the pixel
of interest. Also, the data selecting unit 441 extracts from the
frame #n-1 and frame #n+1 a block centered on the pixel at a
position corresponding to the position of the pixel of interest and
made up of a predetermined number of pixels and multiple blocks
each made up of a predetermined number of pixels from the
surroundings of a pixel at a position corresponding to the pixel of
interest. The data selecting unit 441 supplies the extracted blocks
to the error estimating unit 442.
The error estimating unit 442 detects the correlation between the
block centered on the pixel of interest and the blocks in the
surroundings thereof space-wise or time-wise, supplied from the
data selecting unit 441, and supplies correlation information
indicated the detected correlation to the continuity direction
derivation unit 443. Based on the correlation information from the
error estimating unit 442, the continuity direction derivation unit
443 detects the angle of data continuity in the input image in the
space direction or time direction, corresponding to the lost actual
world 1 light signal continuity, from the position of the block in
the surroundings thereof space-wise or time-wise which has the
greatest correlation, and outputs the data continuity information
which indicates the angle.
Also, the data continuity detecting unit 101 can perform data
continuity detection processing based on component signals of the
input image.
FIG. 105 is a block diagram illustrating the configuration of the
data continuity detecting unit 101 for performing data continuity
detection processing based on component signals of the input
image.
Each of data continuity detecting units 481-1 through 481-3 have
the same configuration as the above-described and or
later-described data continuity detecting unit 101, and executes
the above-described or later-described processing on each component
signals of the input image.
The data continuity detecting unit 481-1 detects the data
continuity based on the first component signal of the input image,
and supplies information indicating the continuity of the data
detected from the first component signal to a determining unit 482.
For example, the data continuity detecting unit 481-1 detects data
continuity based on the brightness signal of the input image, and
supplies information indicating the continuity of the data detected
from the brightness signal to the determining unit 482.
The data continuity detecting unit 481-2 detects the data
continuity based on the second component signal of the input image,
and supplies information indicating the continuity of the data
detected from the second component signal to the determining unit
482. For example, the data continuity detecting unit 481-2 detects
data continuity based on the I signal which is color difference
signal of the input image, and supplies information indicating the
continuity of the data detected from the I signal to the
determining unit 482.
The data continuity detecting unit 481-3 detects the data
continuity based on the third component signal of the input image,
and supplies information indicating the continuity of the data
detected from the third component signal to the determining unit
482. For example, the data continuity detecting unit 481-2 detects
data continuity based on the Q signal which is the color difference
signal of the input image, and supplies information indicating the
continuity of the data detected from the Q signal to the
determining unit 482.
The determining unit 482 detects the final data continuity of the
input image based on the information indicating data continuity
that has been detected from each of the component signals supplied
from the data continuity detecting units 481-1 through 481-3, and
outputs data continuity information indicating the detected data
continuity.
For example, the detecting unit 482 takes as the final data
continuity the greatest data continuity of the data continuities
detected from each of the component signals supplied from the data
continuity detecting units 481-1 through 481-3. Or, for example,
the detecting unit 482 takes as the final data continuity the
smallest data continuity of the data continuities detected from
each of the component signals supplied from the data continuity
detecting units 481-1 through 481-3.
Further, for example, the detecting unit 482 takes as the final
data continuity the average data continuity of the data
continuities detected from each of the component signals supplied
from the data continuity detecting units 481-1 through 481-3. The
determining unit 482 may be arranged so as to taken as the final
data continuity the median (median value) of the data continuities
detected from each of the component signals supplied from the data
continuity detecting units 481-1 through 481-3.
Also, for example, based on signals externally input, the detecting
unit 482 takes as the final data continuity the data continuity
specified by the externally input signals of the data continuities
detected from each of the component signals supplied from the data
continuity detecting units 481-1 through 481-3. The determining
unit 482 may be arranged so as to taken as the final data
continuity a predetermined data continuity of the data continuities
detected from each of the component signals supplied from the data
continuity detecting units 481-1 through 481-3.
Moreover, the detecting unit 482 may be arranged so as to determine
the final data continuity based on the error obtained in the
processing for detecting the data continuity of the component
signals supplied from the data continuity detecting units 481-1
through 481-3. The error which can be obtained in the processing
for data continuity detection will be described later.
FIG. 106 is a diagram illustrating another configuration of the
data continuity detecting unit 101 for performing data continuity
detection based on components signals of the input image.
A component processing unit 491 generates one signal based on the
component signals of the input image, and supplies this to a data
continuity detecting unit 492. For example, the component
processing unit 491 adds values of each of the component signals of
the input image for a pixel at the same position on the screen,
thereby generating a signal made up of the sum of the component
signals.
For example, the component processing unit 491 averages the pixel
values in each of the component signals of the input image with
regard to a pixel at the same position on the screen, thereby
generating a signal made up of the average values of the pixel
values of the component signals.
The data continuity detecting unit 492 detects the data continuity
in the input image, based on the signal input from the component
processing unit 491, and outputs data continuity information
indicating the detected data continuity.
The data continuity detecting unit 492 has the same configuration
as the above-described and or later-described data continuity
detecting unit 101, and executes the above-described or
later-described processing on the signals supplied from the
component processing unit 491.
Thus, the data continuity detecting unit 101 can detect data
continuity by detecting the data continuity of the input image
based on component signals, so the data continuity can be detected
more accurately even in the event that noise and the like is in the
input image. For example, the data continuity detecting unit 101
can detect data continuity angle (gradient), mixture ratio, and
regions having data continuity more precisely, by detecting data
continuity of the input image based on component signals.
Note that the component signals are not restricted to brightness
signals and color difference signals, and may be other component
signals of other formats, such as RGB signals, YUV signals, and so
forth.
As described above, with an arrangement wherein light signals of
the real world are projected, the angle as to the reference axis is
detected of data continuity corresponding to the continuity of real
world light signals that has dropped out from the image data having
continuity of real world light signals of which a part has dropped
out, and the light signals are estimated by estimating the
continuity of the real world light signals that has dropped out
based on the detected angle, processing results which are more
accurate and more precise can be obtained.
Also, with an arrangement wherein multiple sets are extracted of
pixel sets made up of a predetermined number of pixels for each
angle based on a pixel of interest which is the pixel of interest
and the reference axis in image data obtained by light signals of
the real world being projected on multiple detecting elements in
which a part of the continuity of the real world light signals has
dropped out, the correlation of the pixel values of pixels at
corresponding positions in multiple sets which have been extracted
for each angle is detected, the angle of data continuity in the
image data, based on the reference axis, corresponding to the real
world light signal continuity which has dropped out, is detected
based on the detected correlation and the light signals are
estimated by estimating the continuity of the real world light
signals that has dropped out, based on the detected angle of the
data continuity as to the reference axis in the image data,
processing results which are more accurate and more precise as to
the real world events can be obtained.
FIG. 107 is a block diagram illustrating yet another configuration
of the data continuity detecting unit 101.
With the data continuity detecting unit 101 shown in FIG. 107,
light signals of the real world are projected, a region,
corresponding to a pixel of interest which is the pixel of interest
in the image data of which a part of the continuity of the real
world light signals has dropped out, is selected, and a score based
on correlation value is set for pixels wherein the correlation
value of the pixel value of the pixel of interest and the pixel
value of a pixel belonging to a selected region is equal to or
greater than a threshold value, thereby detecting the score of
pixels belonging to the region, and a regression line is detected
based on the detected score, thereby detecting the data continuity
of the image data corresponding to the continuity of the real world
light signals which has dropped out.
Frame memory 501 stores input images in increments of frames, and
supplies the pixel values of the pixels making up stored frames to
a pixel acquiring unit 502. The frame memory 501 can supply pixel
values of pixels of frames of an input image which is a moving
image to the pixel acquiring unit 502, by storing the current frame
of the input image in one page, supplying the pixel values of the
pixel of the frame one frame previous (in the past) as to the
current frame stored in another page to the pixel acquiring unit
502, and switching pages at the switching point-in-time of the
frames of the input image.
The pixel acquiring unit 502 selects the pixel of interest which is
a pixel of interest based on the pixel values of the pixels
supplied from the frame memory 501, and selects a region made up of
a predetermined number of pixels corresponding to the selected
pixel of interest. For example, the pixel acquiring unit 502
selects a region made up of 5.times.5 pixels centered on the pixel
of interest.
The size of the region which the pixel acquiring unit 502 selects
does not restrict the present invention.
The pixel acquiring unit 502 acquires the pixel values of the
pixels of the selected region, and supplies the pixel values of the
pixels of the selected region to a score detecting unit 503.
Based on the pixel values of the pixels of the selected region
supplied from the pixel acquiring unit 502, the score detecting
unit 503 detects the score of pixels belonging to the region, by
setting a score based on correlation for pixels wherein the
correlation value of the pixel value of the pixel of interest and
the pixel value of a pixel belonging to the selected region is
equal to or greater than a threshold value. The details of
processing for setting score based on correlation at the score
detecting unit 503 will be described later.
The score detecting unit 503 supplies the detected score to a
regression line computing unit 504.
The regression line computing unit 504 computes a regression line
based on the score supplied from the score detecting unit 503. For
example, the regression line computing unit 504 computes a
regression line based on the score supplied from the score
detecting unit 503. Also, for example, the regression line
computing unit 504 computes a regression line which is a
predetermined curve, based on the score supplied from the score
detecting unit 503. The regression line computing unit 504 supplies
computation result parameters indicating the computed regression
line and the results of computation to an angle calculating unit
505. The computation results which the computation parameters
indicate include later-described variation and covariation.
The angle calculating unit 505 detects the continuity of the data
of the input image which is image data, corresponding to the
continuity of the light signals of the real world that has dropped
out, based on the regression line indicated by the computation
result parameters supplied from the regression line computing unit
504. For example, based on the regression line indicated by the
computation result parameters supplied from the regression line
computing unit 504, the angle calculating unit 505 detects the
angle of data continuity in the input image based on the reference
axis, corresponding to the dropped actual world 1 light signal
continuity. The angle calculating unit 505 outputs data continuity
information indicating the angle of the data continuity in the
input image based on the reference axis.
The angle of the data continuity in the input image based on the
reference axis will be described with reference to FIG. 108 through
FIG. 110.
In FIG. 108, each circle represents a single pixel, and the double
circle represents the pixel of interest. The colors of the circles
schematically represent the pixel values of the pixels, with the
lighter colors indicating greater pixel values. For example, black
represents a pixel value of 30, while white indicates a pixel value
of 120.
In the event that a person views the image made up of the pixels
shown in FIG. 108, the person who sees the image can recognize that
a straight line is extending in the diagonally upper right
direction.
Upon inputting an input image made up of the pixels shown in FIG.
108, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 107 detects that a straight line is
extending in the diagonally upper right direction.
FIG. 109 is a diagram illustrating the pixel values of the pixels
shown in FIG. 108 with numerical values. Each circle represents one
pixel, and the numerical values in the circles represent the pixel
values.
For example, the pixel value of the pixel of interest is 120, the
pixel value of the pixel above the pixel of interest is 100, and
the pixel value of the pixel below the pixel of interest is 100.
Also, the pixel value of the pixel to the left of the pixel of
interest is 80, and the pixel value of the pixel to the right of
the pixel of interest is 80. In the same way, the pixel value of
the pixel to the lower left of the pixel of interest is 100, and
the pixel value of the pixel to the upper right of the pixel of
interest is 100. The pixel value of the pixel to the upper left of
the pixel of interest is 30, and the pixel value of the pixel to
the lower right of the pixel of interest is 30.
The data continuity detecting unit 101 of which the configuration
is shown in FIG. 107 plots a regression line A as to the input
image shown in FIG. 109, as shown in FIG. 110.
FIG. 111 is a diagram illustrating the relation between change in
pixel values in the input image as to the position of the pixels in
the spatial direction, and the regression line A. The pixel values
of pixels in the region having data continuity change in the form
of a crest, for example, as shown in FIG. 111.
The data continuity detecting unit 101 of which the configuration
is shown in FIG. 107 plots the regression line A by least-square,
weighted with the pixel values of the pixels in the region having
data continuity. The regression line A obtained by the data
continuity detecting unit 101 represents the data continuity in the
neighborhood of the pixel of interest.
The angle of data continuity in the input image based on the
reference axis is detected by obtaining the angle .theta. between
the regression line A and an axis indicating the spatial direction
X which is the reference axis for example, as shown in FIG.
112.
Next, a specific method for calculating the regression line with
the data continuity detecting unit 101 of which the configuration
is shown in FIG. 107.
From the pixel values of pixels in a region made up of 9 pixels in
the spatial direction X and 5 pixels in the spatial direction Y for
a total of 45 pixels, centered on the pixel of interest, supplied
from the pixel acquiring unit 502, for example, the score detecting
unit 503 detects the score corresponding to the coordinates of the
pixels belonging to the region.
For example, the score detecting unit 503 detects the score
L.sub.i,j of the coordinates (x.sub.i, y.sub.j) belonging to the
region, by calculating the score with the computation of Expression
(32).
.function..times..ltoreq.> ##EQU00019##
In Expression (32), P.sub.0,0 represents the pixel value of the
pixel of interest, and P.sub.i,j represents the pixel values of the
pixel at the coordinates (x.sub.i, y.sub.j). Th represents a
threshold value.
i represents the order of the pixel in the spatial direction X in
the region wherein 1.ltoreq.i.ltoreq.k. j represents the order of
the pixel in the spatial direction Y in the region wherein
1.ltoreq.j.ltoreq.l.
k represents the number of pixels in the spatial direction X in the
region, and l represents the number of pixels in the spatial
direction Y in the region. For example, in the event of a region
made up of 9 pixels in the spatial direction X and 5 pixels in the
spatial direction Y for a total of 45 pixels, K is 9 and l is
5.
FIG. 113 is a diagram illustrating an example of a region acquired
by the pixel acquiring unit 502. In FIG. 113, the dotted squares
each represent one pixel.
For example, as shown in FIG. 113, in the event that the region is
made up of 9 pixels centered on the pixel of interest in the
spatial direction X, and is made up of 5 pixels centered on the
pixel of interest in the spatial direction Y, with the coordinates
(x, y) of the pixel of interest being (0, 0), the coordinates (x,
y) of the pixel at the upper left of the region are (-4, 2), the
coordinates (x, y) of the pixel at the upper right of the region
are (4, 2), the coordinates (x, y) of the pixel at the lower left
of the region are (-4, -2), and the coordinates (x, y) of the pixel
at the lower right of the region are (4, -2).
The order i of the pixels at the left side of the region in the
spatial direction X is 1, and the order i of the pixels at the
right side of the region in the spatial direction X is 9. The order
j of the pixels at the lower side of the region in the spatial
direction Y is 1, and the order j of the pixels at the upper side
of the region in the spatial direction Y is 5.
That is to say, with the coordinates (x.sub.5, y.sub.3) of the
pixel of interest as (0, 0), the coordinates (x.sub.1, y.sub.5) of
the pixel at the upper left of the region are (-4, 2), the
coordinates (x.sub.9, y.sub.5) of the pixel at the upper right of
the region are (4, 2), the coordinates (x.sub.1, y.sub.1) of the
pixel at the lower left of the region are (-4, -2), and the
coordinates (x.sub.9, y.sub.1) of the pixel at the lower right of
the region are (4, -2).
The score detecting unit 503 calculates the absolute values of
difference of the pixel value of the pixel of interest and the
pixel values of the pixels belonging to the region as a correlation
value with Expression (32), so this is not restricted to a region
having data continuity in the input image where a fine line image
of the actual world 1 has been projected, rather, score can be
detected representing the feature of spatial change of pixel values
in the region of the input image having two-valued edge data
continuity, wherein an image of an object in the actual world 1
having a straight edge and which is of a monotone color different
from that of the background has been projected.
Note that the score detecting unit 503 is not restricted to the
absolute values of difference of the pixel values of pixels, and
may be arranged to detect the score based on other correlation
values such as correlation coefficients and so forth.
Also, the reason that an exponential function is applied in
Expression (32) is to exaggerate difference in score as to
difference in pixel values, and an arrangement may be made wherein
other functions are applied.
The threshold value Th may be an optional value. For example, the
threshold value Th may be 30.
Thus, the score detecting unit 503 sets a score to pixels having a
correlation value with a pixel value of a pixel belonging to a
selected region, based on the correlation value, and thereby
detects the score of the pixels belonging to the region.
Also, the score detecting unit 503 performs the computation of
Expression (33), thereby calculating the score, whereby the score
L.sub.i,j of the coordinates (x.sub.i, y.sub.j) belonging to the
region is detected.
.ltoreq.> ##EQU00020##
With the score of the coordinates (x.sub.i, y.sub.j) as L.sub.i,j
(1.ltoreq.i.ltoreq.k, 1.ltoreq.j.ltoreq.l), the sum q.sub.i of the
score L.sub.i,j of the coordinate x.sub.i in the spatial direction
Y is expressed by Expression (34), and the sum h.sub.j of the score
L.sub.i,j of the coordinate y.sub.j in the spatial direction X is
expressed by Expression (35).
.times..times. ##EQU00021##
The summation u of the scores is expressed by Expression (36).
.times..times..times..times. ##EQU00022##
In the example shown in FIG. 113, the score L.sub.5,3 of the
coordinate of the pixel of interest is 3, the score L.sub.5,4 of
the coordinate of the pixel above the pixel of interest is 1, the
score L.sub.6,4 of the coordinate of the pixel to the upper right
of the pixel of interest is 4, the score L.sub.6,5 of the
coordinate of the pixel two pixels above and one pixel to the right
of the pixel of interest is 2, and the score L.sub.7,5 of the
coordinate of the pixel two pixels above and two pixels to the
right of the pixel of interest is 3. Also, the score L.sub.5,2 of
the coordinate of the pixel below the pixel of interest is 2, the
score L.sub.4,3 of the coordinate of the pixel to the left of the
pixel of interest is 1, the score L.sub.4,2 of the coordinate of
the pixel to the lower left of the pixel of interest is 3, the
score L.sub.3,2 of the coordinate of the pixel one pixel below and
two pixels to the left of the pixel of interest is 2, and the score
L.sub.3,1 of the coordinate of the pixel two pixels below and two
pixels to the left of the pixel of interest is 4. The score of all
other pixels in the region shown in FIG. 113 is 0, and description
of pixels which have a score of 0 are omitted from FIG. 113.
In the region shown in FIG. 113, the sum q.sub.1 of the scores in
the spatial direction Y is 0, since all scores L wherein i is 1 are
0, and q.sub.2 is 0 since all scores L wherein i is 2 are 0.
q.sub.3 is 6 since L.sub.3,2 is 2 and L.sub.3,1 is 4. In the same
way, q.sub.4 is 4, q.sub.5 is 6, q.sub.6 is 6, q.sub.7 is 3,
q.sub.8 is 0, and q.sub.9 is 0.
In the region shown in FIG. 113, the sum h.sub.1 of the scores in
the spatial direction X is 4, since L.sub.3,1 is 4. h.sub.2 is 7
since L.sub.3,2 is 2, L.sub.4,2 is 3, and L.sub.5,2 is 2. In the
same way, h.sub.3 is 4, h.sub.4 is 5, and h.sub.5 is 5.
In the region shown in FIG. 113, the summation u of scores is
25.
The sum T.sub.x of the results of multiplying the sum q.sub.i of
the scores L.sub.i,j in the spatial direction Y by the coordinate
x.sub.i is shown in Expression (37).
.times..times..times..times..times. ##EQU00023##
The sum T.sub.y of the results of multiplying the sum h.sub.j of
the scores L.sub.i,j in the spatial direction X by the coordinate
y.sub.j is shown in Expression (38).
.times..times..times..times..times. ##EQU00024##
For example, in the region shown in FIG. 113, q.sub.1 is 0 and
x.sub.1 is -4, so q.sub.1 x.sub.1 is 0, and q.sub.2 is 0 and
x.sub.2 is -3, so q.sub.2 x.sub.2 is 0. In the same way, q.sub.3 is
6 and X.sub.3 is -2, so q.sub.3 x.sub.3 is -12; q.sub.4 is 4 and
x.sub.4 is -1, so q.sub.4 x.sub.4 is -4; q.sub.5 is 6 and x.sub.5
is 0, so q.sub.5 x.sub.5 is 0; q.sub.6 is 6 and x.sub.6 is 1, so
q.sub.6 x.sub.6 is 6; q.sub.7 is 3 and x.sub.7 is 2, so q.sub.7
x.sub.7 is 6; q.sub.8 is 0 and x.sub.8 is 3, so q.sub.8 x.sub.8 is
0; and q.sub.9 is 0 and x.sub.9 is 4, so q.sub.9 x.sub.9 is 0.
Accordingly, T.sub.x which is the sum of q.sub.1x.sub.1 through
q.sub.9x.sub.9 is -4.
For example, in the region shown in FIG. 113, h.sub.1 is 4 and
y.sub.1 is -2, so h.sub.1 y.sub.1 is -8, and h.sub.2 is 7 and
y.sub.2 is -1, so h.sub.2 y.sub.2 is -7. In the same way, h.sub.3
is 4 and y.sub.3 is 0, so h.sub.3 y.sub.3 is 0; h.sub.4 is 5 and
y.sub.4 is 1, so h.sub.4y.sub.4 is 5; and h.sub.5 is 5 and y.sub.5
is 2, so h.sub.5y.sub.5 is 10. Accordingly, T.sub.y which is the
sum of h.sub.1y.sub.1 through h.sub.5y.sub.5 is 0.
Also, Q.sub.i is defined as follows.
.times..times. ##EQU00025##
The variation S.sub.x of x is expressed by Expression (40).
.times..times. ##EQU00026##
The variation S.sub.y of y is expressed by Expression (41).
.times..times. ##EQU00027##
The covariation s.sub.xy is expressed by Expression (42).
.times..times..times..times..times..times..times..times.
##EQU00028##
Let us consider obtaining the primary regression line shown in
Expression (43). y=ax+b (43)
The gradient a and intercept b can be obtained as follows by the
least-square method.
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times.
##EQU00029##
However, it should be noted that the conditions necessary for
obtaining a correct regression line is that the scores L.sub.i,j
are distributed in a Gaussian distribution as to the regression
line. To put this the other way around, there is the need for the
score detecting unit 503 to convert the pixel values of the pixels
of the region into the scores L.sub.i,j such that the scores
L.sub.i,j have a Gaussian distribution.
The regression line computing unit 504 performs the computation of
Expression (44) and Expression (45) to obtain the regression
line.
The angle calculating unit 505 performs the computation of
Expression (46) to convert the gradient a of the regression line to
an angle .theta. as to the axis in the spatial direction X, which
is the reference axis. .theta.=tan.sup.-1(a) (46)
Now, in the case of the regression line computing unit 504
computing a regression line which is a predetermined curve, the
angle calculating unit 505 obtains the angle .theta. of the
regression line at the position of the pixel of interest as to the
reference axis.
Here, the intercept b is unnecessary for detecting the data
continuity for each pixel. Accordingly, let us consider obtaining
the primary regression line shown in Expression (47). y=ax (47)
In this case, the regression line computing unit 504 can obtain the
gradient a by the least-square method as in Expression (48).
.times..times..times..times..times..times. ##EQU00030##
The processing for detecting data continuity with the data
continuity detecting unit 101 of which the configuration is shown
in FIG. 107, corresponding to the processing in step S101, will be
described with reference to the flowchart shown in FIG. 114.
In step S501, the pixel acquiring unit 502 selects a pixel of
interest from pixels which have not yet been taken as the pixel of
interest. For example, the pixel acquiring unit 502 selects the
pixel of interest in raster scan order. In step S502, the pixel
acquiring unit 502 acquires the pixel values of the pixel contained
in a region centered on the pixel of interest, and supplies the
pixel values of the pixels acquired to the score detecting unit
503. For example, the pixel acquiring unit 502 selects a region
made up of 9.times.5 pixels centered on the pixel of interest, and
acquires the pixel values of the pixels contained in the
region.
In step S503, the score detecting unit 503 converts the pixel
values of the pixels contained in the region into scores, thereby
detecting scores. For example, the score detecting unit 503
converts the pixel values into scores L.sub.i,j by the computation
shown in Expression (32). In this case, the score detecting unit
503 converts the pixel values of the pixels of the region into the
scores L.sub.i,j such that the scores L.sub.i,j have a Gaussian
distribution. The score detecting unit 503 supplies the converted
scores to the regression line computing unit 504.
In step S504, the regression line computing unit 504 obtains a
regression line based on the scores supplied from the score
detecting unit 503. For example, the regression line computing unit
504 obtains the regression line based on the scores supplied from
the score detecting unit 503. More specifically, the regression
line computing unit 504 obtains the regression line by executing
the computation shown in Expression (44) and Expression (45). The
regression line computing unit 504 supplies computation result
parameters indicating the regression line which is the result of
computation, to the angle calculating unit 505.
In step S505, the angle calculating unit 505 calculates the angle
of the regression line as to the reference axis, thereby detecting
the data continuity of the image data, corresponding to the
continuity of the light signals of the real world that has dropped
out. For example, the angle calculating unit 505 converts the
gradient a of the regression line into the angle .theta. as to the
axis of the spatial direction X which is the reference axis, by the
computation of Expression (46).
Note that an arrangement may be made wherein the angle calculating
unit 505 outputs data continuity information indicating the
gradient a.
In step S506, the pixel acquiring unit 502 determines whether or
not the processing of all pixels has ended, and in the event that
determination is made that the processing of all pixels has not
ended, the flow returns to step S501, a pixel of interest is
selected from the pixels which have not yet been taken as a pixel
of interest, and the above-described processing is repeated.
In the event that determination is made in step S506 that the
processing of all pixels has ended, the processing ends.
Thus, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 107 can detect the angle of data
continuity in the image data based on the reference axis,
corresponding to the dropped continuity of the actual world 1 light
signals.
Particularly, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 107 can obtain angles smaller than
pixels, based on the pixel values of pixels in a relatively narrow
region.
As described above, in a case wherein light signals of the real
world are projected, a region, corresponding to a pixel of interest
which is the pixel of interest in the image data of which a part of
the continuity of the real world light signals has dropped out, is
selected, and a score based on correlation value is set for pixels
wherein the correlation value of the pixel value of the pixel of
interest and the pixel value of a pixel belonging to a selected
region is equal to or greater than a threshold value, thereby
detecting the score of pixels belonging to the region, and a
regression line is detected based on the detected score, thereby
detecting the data continuity of the image data corresponding to
the continuity of the real world light signals which has dropped
out, and subsequently estimating the light signals by estimating
the continuity of the dropped real world light signal based on the
detected data of the image data, processing results which are more
accurate and more precise as to events in the real world can be
obtained.
Note that with the data continuity detecting unit 101 of which the
configuration is shown in FIG. 107, an arrangement wherein the
pixel values of pixels in a predetermined region of the frame of
interest where the pixel of interest belongs and in frames before
and after the frame of interest time-wise are converted into
scores, and a regression plane is obtained based on the scores,
allows the angle of time-directional data continuity to be detected
along with the angle of the data continuity in the spatial
direction.
FIG. 115 is a block diagram illustrating yet another configuration
of the data continuity detecting unit 101.
With the data continuity detecting unit 101 shown in FIG. 115,
light signals of the real world are projected, a region,
corresponding to a pixel of interest which is the pixel of interest
in the image data of which a part of the continuity of the real
world light signals has dropped out, is selected, and a score based
on correlation value is set for pixels wherein the correlation
value of the pixel value of the pixel of interest and the pixel
value of a pixel belonging to a selected region is equal to or
greater than a threshold value, thereby detecting the score of
pixels belonging to the region, and a regression line is detected
based on the detected score, thereby detecting the data continuity
of the image data corresponding to the continuity of the real world
light signals which has dropped out.
Frame memory 601 stores input images in increments of frames, and
supplies the pixel values of the pixels making up stored frames to
a pixel acquiring unit 602. The frame memory 601 can supply pixel
values of pixels of frames of an input image which is a moving
image to the pixel acquiring unit 602, by storing the current frame
of the input image in one page, supplying the pixel values of the
pixel of the frame one frame previous (in the past) as to the
current frame stored in another page to the pixel acquiring unit
602, and switching pages at the switching point-in-time of the
frames of the input image.
The pixel acquiring unit 602 selects the pixel of interest which is
a pixel of interest based on the pixel values of the pixels
supplied from the frame memory 601, and selects a region made up of
a predetermined number of pixels corresponding to the selected
pixel of interest. For example, the pixel acquiring unit 602
selects a region made up of 5.times.5 pixels centered on the pixel
of interest.
The size of the region which the pixel acquiring unit 602 selects
does not restrict the present invention.
The pixel acquiring unit 602 acquires the pixel values of the
pixels of the selected region, and supplies the pixel values of the
pixels of the selected region to a score detecting unit 603.
Based on the pixel values of the pixels of the selected region
supplied from the pixel acquiring unit 602, the score detecting
unit 603 detects the score of pixels belonging to the region, by
setting a score based on correlation value for pixels wherein the
correlation value of the pixel value of the pixel of interest and
the pixel value of a pixel belonging to the selected region is
equal to or greater than a threshold value. The details of
processing for setting score based on correlation at the score
detecting unit 603 will be described later.
The score detecting unit 603 supplies the detected score to a
regression line computing unit 604.
The regression line computing unit 604 computes a regression line
based on the score supplied from the score detecting unit 603. For
example, the regression line computing unit 604 computes a
regression line based on the score supplied from the score
detecting unit 603. Also, for example, the regression line
computing unit 604 computes a regression line which is a
predetermined curve, based on the score supplied from the score
detecting unit 603. The regression line computing unit 604 supplies
computation result parameters indicating the computed regression
line and the results of computation to an region calculating unit
605. The computation results which the computation parameters
indicate include later-described variation and covariation.
The region calculating unit 605 detects the region having the
continuity of the data of the input image which is image data,
corresponding to the continuity of the light signals of the real
world that has dropped out, based on the regression line indicated
by the computation result parameters supplied from the regression
line computing unit 604.
FIG. 116 is a diagram illustrating the relation between change in
pixel values in the input image as to the position of the pixels in
the spatial direction, and the regression line A. The pixel values
of pixels in the region having data continuity change in the form
of a crest, for example, as shown in FIG. 116.
The data continuity detecting unit 101 of which the configuration
is shown in FIG. 115 plots the regression line A by least-square,
weighted with the pixel values of the pixels in the region having
data continuity. The regression line A obtained by the data
continuity detecting unit 101 represents the data continuity in the
neighborhood of the pixel of interest.
Plotting a regression line means approximation assuming a Gaussian
function. As shown in FIG. 117, the data continuity detecting unit
of which the configuration is illustrated in FIG. 115 can tell the
general width of the region in the data 3 where the image of the
fine line has been projected, by obtaining standard deviation, for
example. Also, the data continuity detecting unit of which the
configuration is illustrated in FIG. 115 can tell the general width
of the region in the data 3 where the image of the fine line has
been projected, based on correlation coefficients.
Next, description will be made regarding a specific method for
calculating the regression line with the data continuity detecting
unit 101 of which the configuration is shown in FIG. 115.
From the pixel values of pixels in a region made up of 9 pixels in
the spatial direction X and 5 pixels in the spatial direction Y for
a total of 45 pixels, centered on the pixel of interest, supplied
from the pixel acquiring unit 602, for example, the score detecting
unit 603 detects the score corresponding to the coordinates of the
pixels belonging to the region.
For example, the score detecting unit 603 detects the score
L.sub.i,j of the coordinates (x.sub.i, y.sub.j) belonging to the
region, by calculating the score with the computation of Expression
(49).
.function..times..ltoreq.> ##EQU00031##
In Expression (49), P.sub.0,0 represents the pixel value of the
pixel of interest, and P.sub.i,j represents the pixel values of the
pixel at the coordinates (x.sub.i, y.sub.j). Th represents the
threshold value.
i represents the order of the pixel in the spatial direction X in
the region wherein 1.ltoreq.i.ltoreq.k. j represents the order of
the pixel in the spatial direction Y in the region wherein
1.ltoreq.j.ltoreq.l.
k represents the number of pixels in the spatial direction X in the
region, and l represents the number of pixels in the spatial
direction Y in the region. For example, in the event of a region
made up of 9 pixels in the spatial direction X and 5 pixels in the
spatial direction Y for a total of 45 pixels, K is 9 and l is
5.
FIG. 118 is a diagram illustrating an example of a region acquired
by the pixel acquiring unit 602. In FIG. 118, the dotted squares
each represent one pixel.
For example, as shown in FIG. 118, in the event that the region is
made up of 9 pixels centered on the pixel of interest in the
spatial direction X, and is made up of 5 pixels centered on the
pixel of interest in the spatial direction Y, with the coordinates
(x, y) of the pixel of interest being (0, 0), the coordinates (x,
y) of the pixel at the upper left of the region are (-4, 2), the
coordinates (x, y) of the pixel at the upper right of the region
are (4, 2), the coordinates (x, y) of the pixel at the lower left
of the region are (-4, -2), and the coordinates (x, y) of the pixel
at the lower right of the region are (4, -2).
The order i of the pixels at the left side of the region in the
spatial direction X is 1, and the order i of the pixels at the
right side of the region in the spatial direction X is 9. The order
j of the pixels at the lower side of the region in the spatial
direction Y is 1, and the order j of the pixels at the upper side
of the region in the spatial direction Y is 5.
That is to say, with the coordinates (x.sub.5, y.sub.3) of the
pixel of interest as (0, 0), the coordinates (x.sub.1, y.sub.5) of
the pixel at the upper left of the region are (-4, 2), the
coordinates (x.sub.9, y.sub.5) of the pixel at the upper right of
the region are (4, 2), the coordinates (x.sub.1, y.sub.1) of the
pixel at the lower left of the region are (-4, -2), and the
coordinates (x.sub.9, y.sub.1) of the pixel at the lower right of
the region are (4, -2).
The score detecting unit 603 calculates the absolute values of
difference of the pixel value of the pixel of interest and the
pixel values of the pixels belonging to the region as a correlation
value with Expression (49), so this is not restricted to a region
having data continuity in the input image where a fine line image
of the actual world 1 has been projected, rather, score can be
detected representing the feature of spatial change of pixel values
in the region of the input image having two-valued edge data
continuity, wherein an image of an object in the actual world 1
having a straight edge and which is of a monotone color different
from that of the background has been projected.
Note that the score detecting unit 603 is not restricted to the
absolute values of difference of the pixel values of the pixels,
and may be arranged to detect the score based on other correlation
values such as correlation coefficients and so forth.
Also, the reason that an exponential function is applied in
Expression (49) is to exaggerate difference in score as to
difference in pixel values, and an arrangement may be made wherein
other functions are applied.
The threshold value Th may be an optional value. For example, the
threshold value Th may be 30.
Thus, the score detecting unit 603 sets a score to pixels having a
correlation value with a pixel value of a pixel belonging to a
selected region equal to or greater than the threshold value, based
on the correlation value, and thereby detects the score of the
pixels belonging to the region.
Also, the score detecting unit 603 performs the computation of
Expression (50) for example, thereby calculating the score, whereby
the score L.sub.i,j of the coordinates (x.sub.i, y.sub.j) belonging
to the region is detected.
.ltoreq.> ##EQU00032##
With the score of the coordinates (x.sub.i, y.sub.j) as L.sub.i,j
(1.ltoreq.i.ltoreq.k, 1.ltoreq.j.ltoreq.l), the sum q.sub.i of the
score L.sub.i,j of the coordinate x.sub.i in the spatial direction
Y is expressed by Expression (51), and the sum h.sub.j of the score
L.sub.i,j of the coordinate y.sub.j in the spatial direction X is
expressed by Expression (52).
.times..times. ##EQU00033##
The summation u of the scores is expressed by Expression (53).
.times..times..times..times. ##EQU00034##
In the example shown in FIG. 118, the score L.sub.5,3 of the
coordinate of the pixel of interest is 3, the score L.sub.5,4 of
the coordinate of the pixel above the pixel of interest is 1, the
score L.sub.6,4 of the coordinate of the pixel to the upper right
of the pixel of interest is 4, the score L.sub.6,5 of the
coordinate of the pixel two pixels above and one pixel to the right
of the pixel of interest is 2, and the score L.sub.7,5 of the
coordinate of the pixel two pixels above and two pixels to the
right of the pixel of interest is 3. Also, the score L.sub.5,2 of
the coordinate of the pixel below the pixel of interest is 2, the
score L.sub.4,3 of the coordinate of the pixel to the left of the
pixel of interest is 1, the score L.sub.4,2 of the coordinate of
the pixel to the lower left of the pixel of interest is 3, the
score L.sub.4,2 of the coordinate of the pixel one pixel below and
two pixels to the left of the pixel of interest is 2, and the score
L.sub.3,1 of the coordinate of the pixel two pixels below and two
pixels to the left of the pixel of interest is 4. The score of all
other pixels in the region shown in FIG. 118 is 0, and description
of pixels which have a score of 0 are omitted from FIG. 118.
In the region shown in FIG. 118, the sum q.sub.1 of the scores in
the spatial direction Y is 0, since all scores L wherein i is 1 are
0, and q.sub.2 is 0 since all scores L wherein i is 2 are 0.
q.sub.3 is 6 since L.sub.3,2 is 2 and L.sub.3,1 is 4. In the same
way, q.sub.4 is 4, q.sub.5 is 6, q.sub.6 is 6, q.sub.7 is 3,
q.sub.8 is 0, and q.sub.9 is 0.
In the region shown in FIG. 118, the sum h.sub.1 of the scores in
the spatial direction X is 4, since L.sub.3,1 is 4. h.sub.2 is 7
since L.sub.3,2 is 2, L.sub.4,2 is 3, and L.sub.5,2 is 2. In the
same way, h.sub.3 is 4, h.sub.4 is 5, and h.sub.5 is 5.
In the region shown in FIG. 118, the summation u of scores is
25.
The sum T.sub.x of the results of multiplying the sum q.sub.i of
the scores L.sub.i,j in the spatial direction Y by the coordinate
x.sub.i is shown in Expression (54).
.times..times..times..times..times. ##EQU00035##
The sum T.sub.y of the results of multiplying the sum h.sub.j of
the scores L.sub.i,j in the spatial direction X by the coordinate
y.sub.j is shown in Expression (55).
.times..times..times..times..times. ##EQU00036##
For example, in the region shown in FIG. 118, q.sub.1 is 0 and
x.sub.1 is -4, so q.sub.1 x.sub.1 is 0, and q.sub.2 is 0 and
x.sub.2 is -3, so q.sub.2 x.sub.2 is 0. In the same way, q.sub.3 is
6 and x.sub.3 is -2, so q.sub.3 x.sub.3 is -12; q.sub.4 is 4 and
x.sub.4 is -1, so q.sub.4 x.sub.4 is -4; q.sub.5 is 6 and x.sub.5
is 0, so q.sub.5 x.sub.5 is 0; q.sub.6 is 6 and x.sub.6 is 1, so
q.sub.6 x.sub.6 is 6; q.sub.7 is 3 and x.sub.7 is 2, so q.sub.7
x.sub.7 is 6; q.sub.8 is 0 and x.sub.8 is 3, so q.sub.8 x.sub.8 is
0; and q.sub.9 is 0 and x.sub.9 is 4, so q.sub.9 x.sub.9 is 0.
Accordingly, T.sub.x which is the sum of q.sub.1x.sub.1 through
q.sub.9x.sub.9 is -4.
For example, in the region shown in FIG. 118, h.sub.1 is 4 and
y.sub.1 is -2, so h.sub.1 y.sub.1 is -8, and h.sub.2 is 7 and
y.sub.2 is -1, so h.sub.2 y.sub.2 is -7. In the same way, h.sub.3
is 4 and y.sub.3 is 0, so h.sub.3 y.sub.3 is 0; h.sub.4 is 5 and
y.sub.4 is 1, so h.sub.4y.sub.4 is 5; and h.sub.5 is 5 and y.sub.5
is 2, so h.sub.5y.sub.5 is 10. Accordingly, T.sub.y which is the
sum of h.sub.1y.sub.1 through h.sub.5y.sub.5 is 0.
Also, Q.sub.i is defined as follows.
.times..times. ##EQU00037##
The variation S.sub.x of x is expressed by Expression (57).
.times..times. ##EQU00038##
The variation S.sub.y of y is expressed by Expression (58).
.times..times. ##EQU00039##
The covariation s.sub.xy is expressed by Expression (59).
.times..times..times..times..times..times..times..times.
##EQU00040##
Let us consider obtaining the primary regression line shown in
Expression (60). y=ax+b (60)
The gradient a and intercept b can be obtained as follows by the
least-square method.
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times. ##EQU00041##
However, it should be noted that the conditions necessary for
obtaining a correct regression line is that the scores L.sub.i,j
are distributed in a Gaussian distribution as to the regression
line. To put this the other way around, there is the need for the
score detecting unit 603 to convert the pixel values of the pixels
of the region into the scores L.sub.i,j such that the scores
L.sub.i,j have a Gaussian distribution.
The regression line computing unit 604 performs the computation of
Expression (61) and Expression (62) to obtain the regression
line.
Also, the intercept b is unnecessary for detecting the data
continuity for each pixel. Accordingly, let us consider obtaining
the primary regression line shown in Expression (63). y=ax (63)
In this case, the regression line computing unit 604 can obtain the
gradient a by the least-square method as in Expression (64).
.times..times..times..times..times..times. ##EQU00042##
With a first technique for determining the region having data
continuity, the estimation error of the regression line shown in
Expression (60) is used.
The variation S.sub.yx of y is obtained with the computation shown
in Expression (65).
.times. ##EQU00043##
Scattering of the estimation error is obtained by the computation
shown in Expression (66) using variation.
##EQU00044##
Accordingly, the following Expression yields the standard
deviation.
##EQU00045##
However, in the case of handling a region where a fine line image
has been projected, the standard deviation is an amount worth the
width of the fine line, so determination cannot be categorically
made that great standard deviation means that a region is not the
region with data continuity. However, for example, information
indicating detected regions using standard deviation can be
utilized to detect regions where there is a great possibility that
class classification adaptation processing breakdown will occur,
since class classification adaptation processing breakdown occurs
at portions of the region having data continuity where the fine
line is narrow.
The region calculating unit 605 calculates the standard deviation
by the computation shown in Expression (67), and calculates the
region of the input image having data continuity, based on the
standard deviation, for example. The region calculating unit 605
multiplies the standard deviation by a predetermined coefficient so
as to obtain distance, and takes the region within the obtained
distance from the regression line as a region having data
continuity. For example, the region calculating unit 605 calculates
the region within the standard deviation distance from the
regression line as a region having data continuity, with the
regression line as the center thereof.
With a second technique, the correlation of score is used for
detecting a region having data continuity.
The correlation coefficient r.sub.xy can be obtained by the
computation shown in Expression (68), based on the variation
S.sub.x of x, the variation S.sub.y of y, and the covariation
S.sub.xy.
.times. ##EQU00046##
Correlation includes positive correlation and negative correlation,
so the region calculating unit 605 obtains the absolute value of
the correlation coefficient r.sub.xy, and determines that the
closer to 1 the absolute value of the correlation coefficient
r.sub.xy is, the greater the correlation is. More specifically, the
region calculating unit 605 compares the threshold value with the
absolute value of the correlation coefficient r.sub.xy, and detects
a region wherein the correlation coefficient r.sub.xy is equal to
or greater than the threshold value as a region having data
continuity.
The processing for detecting data continuity with the data
continuity detecting unit 101 of which the configuration is shown
in FIG. 115, corresponding to the processing in step S101, will be
described with reference to the flowchart shown in FIG. 119.
In step S601, the pixel acquiring unit 602 selects a pixel of
interest from pixels which have not yet been taken as the pixel of
interest. For example, the pixel acquiring unit 602 selects the
pixel of interest in raster scan order. In step S602, the pixel
acquiring unit 602 acquires the pixel values of the pixel contained
in a region centered on the pixel of interest, and supplies the
pixel values of the pixels acquired to the score detecting unit
603. For example, the pixel acquiring unit 602 selects a region
made up of 9.times.5 pixels centered on the pixel of interest, and
acquires the pixel values of the pixels contained in the
region.
In step S603, the score detecting unit 603 converts the pixel
values of the pixels contained in the region into scores, thereby
detecting scores. For example, the score detecting unit 603
converts the pixel values into scores L.sub.i,j by the computation
shown in Expression (49). In this case, the score detecting unit
603 converts the pixel values of the pixels of the region into the
scores L.sub.i,j such that the scores L.sub.i,j have a Gaussian
distribution. The score detecting unit 603 supplies the converted
scores to the regression line computing unit 604.
In step S604, the regression line computing unit 604 obtains a
regression line based on the scores supplied from the score
detecting unit 603. For example, the regression line computing unit
604 obtains the regression line based on the scores supplied from
the score detecting unit 603. More specifically, the regression
line computing unit 604 obtains the regression line by executing
the computation shown in Expression (61) and Expression (62). The
regression line computing unit 604 supplies computation result
parameters indicating the regression line which is the result of
computation, to the region calculating unit 605.
In step S605, the region calculating unit 605 calculates the
standard deviation regarding the regression line. For example, an
arrangement may be made wherein the region calculating unit 605
calculates the standard deviation as to the regression line by the
computation of Expression (67).
In step S606, the region calculating unit 605 determines the region
of the input image having data continuity, from the standard
deviation. For example, the region calculating unit 605 multiplies
the standard deviation by a predetermined coefficient to obtain
distance, and determines the region within the obtained distance
from the regression line to be the region having data
continuity.
The region calculating unit 605 outputs data continuity information
indicating a region having data continuity.
In step S607, the pixel acquiring unit 602 determines whether or
not the processing of all pixels has ended, and in the event that
determination is made that the processing of all pixels has not
ended, the flow returns to step S601, a pixel of interest is
selected from the pixels which have not yet been taken as a pixel
of interest, and the above-described processing is repeated.
In the event that determination is made in step S607 that the
processing of all pixels has ended, the processing ends.
Other processing for detecting data continuity with the data
continuity detecting unit 101 of which the configuration is shown
in FIG. 115, corresponding to the processing in step S101, will be
described with reference to the flowchart shown in FIG. 120. The
processing of step S621 through step S624 is the same as the
processing of step S601 through step S604, so description thereof
will be omitted.
In step S625, the region calculating unit 605 calculates a
correlation coefficient regarding the regression line. For example,
the region calculating unit 605 calculates the correlation
coefficient as to the regression line by the computation of
Expression (68).
In step S626, the region calculating unit 605 determines the region
of the input image having data continuity, from the correlation
coefficient. For example, the region calculating unit 605 compares
the absolute value of the correlation coefficient with a threshold
value stored beforehand, and determines a region wherein the
absolute value of the correlation coefficient is equal to or
greater than the threshold value to be the region having data
continuity.
The region calculating unit 605 outputs data continuity information
indicating a region having data continuity.
The processing of step S627 is the same as the processing of step
S607, so description thereof will be omitted.
Thus, the data continuity detecting unit 101 of which the
configuration is shown in FIG. 115 can detect the region in the
image data having data continuity, corresponding to the dropped
actual world 1 light signal continuity.
As described above, in a case wherein light signals of the real
world are projected, a region, corresponding to a pixel of interest
which is the pixel of interest in the image data of which a part of
the continuity of the real world light signals has dropped out, is
selected, and a score based on correlation value is set for pixels
wherein the correlation value of the pixel value of the pixel of
interest and the pixel value of a pixel belonging to a selected
region is equal to or greater than a threshold value, thereby
detecting the score of pixels belonging to the region, and a
regression line is detected based on the detected score, thereby
detecting the region having the data continuity of the image data
corresponding to the continuity of the real world light signals
which has dropped out, and subsequently estimating the light
signals by estimating the dropped real world light signal
continuity based on the detected data continuity of the image data,
processing results which are more accurate and more precise as to
events in the real world can be obtained.
FIG. 121 illustrates the configuration of another form of the data
continuity detecting unit 101.
The data continuity detecting unit 101 shown in FIG. 121 comprises
a data selecting unit 701, a data supplementing unit 702, and a
continuity direction derivation unit 703.
The data selecting unit 701 takes each pixel of the input image as
the pixel of interest, selects pixel value data of pixels
corresponding to each pixel of interest, and outputs this to the
data supplementing unit 702.
The data supplementing unit 702 performs least-square
supplementation computation based on the data input from the data
selecting unit 701, and outputs the supplementation computation
results of the continuity direction derivation unit 703. The
supplementation computation by the data supplementing unit 702 is
computation regarding the summation item used in the
later-described least-square computation, and the computation
results thereof can be said to be the feature of the image data for
detecting the angle of continuity.
The continuity direction derivation unit 703 computes the
continuity direction, i.e., the angle as to the reference axis
which the data continuity has (e.g., the gradient or direction of a
fine line or two-valued edge) from the supplementation computation
results input by the data supplementing unit 702, and outputs this
as data continuity information.
Next, the overview of the operations of the data continuity
detecting unit 101 in detecting continuity (direction or angle)
will be described with reference to FIG. 122. Portions in FIG. 122
and FIG. 123 which correspond with those in FIG. 6 and FIG. 7 are
denoted with the same symbols, and description thereof in the
following will be omitted as suitable.
As shown in FIG. 122, signals of the actual world 1 (e.g., an
image), are imaged on the photoreception face of a sensor 2 (e.g.,
a CCD (Charge Coupled Device) or CMOS (Complementary Metal-Oxide
Semiconductor)), by an optical system 141 (made up of lenses, an
LPF (Low Pass Filter), and the like, for example). The sensor 2 is
configured of a device having integration properties, such as a CCD
or CMOS, for example. Due to this configuration, the image obtained
from the data 3 output from the sensor 2 is an image differing from
the image of the actual world 1 (difference as to the image of the
actual world 1 occurs).
Accordingly, as shown in FIG. 123, the data continuity detecting
unit 101 uses a model 705 to describe in an approximate manner the
actual world 1 by an approximation expression and extracts the data
continuity from the approximation expression. The model 705 is
represented by, for example, N variables. More accurately, the
model 705 approximates (describes) signals of the actual world
1.
In order to predict the model 705, the data continuity detecting
unit 101 extracts M pieces of data 706 from the data 3.
Consequently, the model 705 is constrained by the continuity of the
data.
That is to say, the model 705 approximates continuity of the
(information (signals) indicating) events of the actual world 1
having continuity (constant characteristics in a predetermined
dimensional direction), which generates the data continuity in the
data 3 when obtained with the sensor 2.
Now, in the event that the number M of the data 706 is N, which is
the number N of variables of the model 705, or more, the model 705
represented by the N variables can be predicted from M pieces of
data 706.
Further, by predicting the model 705 approximating (describing) the
signals of) the actual world 1, the data continuity detecting unit
101 derives the data continuity contained in the signals which are
information of the actual world 1 as, for example, fine line or
two-valued edge direction (the gradient, or the angle as to an axis
in a case wherein a predetermined direction is taken as an axis),
and outputs this as data continuity information.
Next, the data continuity detecting unit 101 which outputs the
direction (angle) of a fine line from the input image as data
continuity information will be described with reference to FIG.
124.
The data selecting unit 701 is configured of a horizontal/vertical
determining unit 711, and a data acquiring unit 712. The
horizontal/vertical determining unit 711 determines, from the
difference in pixel values between the pixel of interest and the
surrounding pixels, whether the angle as to the horizontal
direction of the fine line in the input image is a fine line closer
to the horizontal direction or is a fine line closer to the
vertical direction, and outputs the determination results to the
data acquiring unit 712 and data supplementing unit 702.
In more detail, for example, in the sense of this technique, other
techniques may be used as well. For example, simplified
16-directional detection may be used. As shown in FIG. 125, of the
difference between the pixel of interest and the surrounding pixels
(difference in pixel values between the pixels), the
horizontal/vertical determining unit 711 obtains the difference
between the sum of difference (activity) between pixels in the
horizontal direction (hdiff) and the sum of difference (activity)
between pixels in the vertical direction (vdiff), and determines
whether the sum of difference is greater between the pixel of
interest and pixels adjacent thereto in the vertical direction, or
between the pixel of interest and pixels adjacent thereto in the
horizontal direction. Now, in FIG. 125, each grid represents a
pixel, and the pixel at the center of the diagram is the pixel of
interest. Also, the differences between pixels indicated by the
dotted arrows in the diagram are the differences between pixels in
the horizontal direction, and the sum thereof is indicated by
hdiff. Also, the differences between pixels indicated by the solid
arrows in the diagram are the differences between pixels in the
vertical direction, and the sum thereof is indicated by vdiff.
Based on the sum of differences hdiff of the pixel values of the
pixels in the horizontal direction, and the sum of differences
vdiff of the pixel values of the pixels in the vertical direction,
that have been thus obtained, in the event that (hdiff minus vdiff)
is positive, this means that the change (activity) of pixel values
between pixels is greater in the horizontal direction than the
vertical direction, so in a case wherein the angle as to the
horizontal direction is represented by .theta. (0 degrees degrees
.ltoreq..theta..ltoreq.180 degrees degrees) as shown in FIG. 126,
the horizontal/vertical determining unit 711 determines that the
pixels belong to a fine line which is 45 degrees
degrees<.theta..ltoreq.135 degrees degrees, i.e., an angle
closer to the vertical direction, and conversely, in the event that
this is negative, this means that the change (activity) of pixel
values between pixels is greater in the vertical direction, so the
horizontal/vertical determining unit 711 determines that the pixels
belong to a fine line which is 0 degrees
degrees.ltoreq..theta.<45 degrees degrees or 135 degrees
degrees<.theta..ltoreq.180 degrees degrees, i.e., an angle
closer to the horizontal direction (pixels in the direction (angle)
in which the fine line extends each are pixels representing the
fine line, so change (activity) between those pixels should be
smaller).
Also, the horizontal/vertical determining unit 711 has a counter
(not shown) for identifying individual pixels of the input image,
and can be used whenever suitable or necessary.
Also, while description has been made in FIG. 125 regarding an
example of comparing the sum of difference of pixel values between
pixels in the vertical direction and horizontal direction in a 3
pixel.times.3 pixel range centered on the pixel of interest, to
determine whether the fine line is closer to the vertical direction
or closer to the horizontal direction, but the direction of the
fine line can be determined with the same technique using a greater
number of pixels, for example, determination may be made based on
blocks of 5 pixels.times.5 pixels centered on the pixel of
interest, 7 pixels.times.7 pixels, and so forth, i.e., a greater
number of pixels.
Based on the determination results regarding the direction of the
fine line input from the horizontal/vertical determining unit 711,
the data acquiring unit 712 reads out (acquires) pixel values in
increments of blocks made up of multiple pixels arrayed in the
horizontal direction corresponding to the pixel of interest, or in
increments of blocks made up of multiple pixels arrayed in the
vertical direction, and along with data of difference between
pixels adjacent in the direction according to the determination
results from the horizontal/vertical determining unit 711 between
multiple corresponding pixels for each pixel of interest read out
(acquired), maximum value and minimum value data of pixel values of
the pixels contained in blocks of a predetermined number of pixels
is output to the data supplementing unit 702. Hereafter, a block
made up of multiple pixels obtained corresponding to the pixel of
interest by the data acquiring unit 712 will be referred to as an
acquired block (of the multiple pixels (each represented by a grid)
shown in FIG. 139 described later for example, with the pixel
indicated by the black square as the pixel of interest, an acquired
block is the three pixels above and below, and one pixel to the
right and left, for a total of 15 pixels.
The difference supplementing unit 721 of the data supplementing
unit 702 detects the difference data input from the data selecting
unit 701, executes supplementing processing necessary for solution
of the later-described least-square method, based on the
determination results of horizontal direction or vertical direction
input from the horizontal/vertical determining unit 711 of the data
selecting unit 701, and outputs the supplementing results to the
continuity direction derivation unit 703. More specifically, of the
multiple pixels, the data of difference in the pixel values between
the pixel i adjacent in the direction determined by the
horizontal/vertical determining unit 711 and the pixel (i+1) is
taken as yi, and in the event that the acquired block corresponding
to the pixel of interest is made up of n pixels, the difference
supplementing unit 721 computes supplementing of
(y1).sup.2+(y2).sup.2+(y3).sup.2+ . . . for each horizontal
direction or vertical direction, and outputs to the continuity
direction derivation unit 703.
Upon obtaining the maximum value and minimum value of pixel values
of pixels contained in a block set for each of the pixels contained
in the acquired block corresponding to the pixel of interest input
from the data selecting unit 701 (hereafter referred to as a
dynamic range block (of the pixels in the acquired block indicated
in FIG. 139 which will be described later, a dynamic range block of
the three pixels above and below the pixel pixl2 for a total of 7
pixels, illustrated as the dynamic range block B1 surrounded with a
black solid line)), a MaxMin acquiring unit 722 computes (detects)
from the difference thereof a dynamic range Dri (the difference
between the maximum value and minimum value of pixel values of
pixels contained in the dynamic range block corresponding to the
i'th pixel in the acquired block), and outputs this to a difference
supplementing unit 723.
The difference supplementing unit 723 detects the dynamic range Dri
input from the MaxMin acquiring unit 722 and the difference data
input from the data selecting unit 701, supplements each horizontal
direction or vertical direction input from the horizontal/vertical
determining unit 711 of the data selecting unit 701 with a value
obtained by multiplying the dynamic range Dri and the difference
data yi based on the dynamic range Dri and the difference data
which have been detected, and outputs the computation results to
the continuity direction derivation unit 703. That is to say, the
computation results which the difference supplementing unit 723
outputs is y1.times.Dr1+y2.times.Dr2+y3.times.Dr3+ . . . in each
horizontal direction or vertical direction.
The continuity direction computation unit 731 of the continuity
direction derivation unit 703 computes the angle (direction) of the
fine line based on the supplemented computation results in each
horizontal direction or vertical direction input from the data
supplementing unit 702, and outputs the computed angle as
continuity information.
Now, the method for computing the direction (gradient or angle of
the fine line) of the fine line will be described.
Enlarging the portion surrounded by the white line in an input
image such as shown in FIG. 127A shows that the fine line (the
white line extending diagonally in the upwards right direction in
the drawing) is actually displayed as in FIG. 127B. That is to say,
in the real world, the image is such that as shown in FIG. 127C,
the two levels of fine-line level (the lighter hatched portion in
FIG. 127C) and the background level form boundaries, and no other
levels exist. Conversely, the image taken with the sensor 2, i.e.,
the image imaged in increments of pixels, is an image wherein, as
shown in FIG. 127B, there is a repeated array in the fine line
direction of blocks which are made up of multiple pixels with the
background level and the fine line level spatially mixed due to the
integration effects, arrayed in the vertical direction so that the
ratio (mixture ratio) thereof changes according to a certain
pattern. Note that in FIG. 127B, each square-shaped grid represents
one pixel of the CCD, and we will say that the length of each side
thereof is d_CCD. Also, the portions of the grids filled in
lattice-like are the minimum value of the pixel values, equivalent
to the background level, and the other portions filled in hatched
have a greater pixel value the less dense the shading is
(accordingly, white grids with no shading have the maximum value of
the pixel values).
In the event that a fine line exists on the background in the real
world as shown in FIG. 128A, the image of the real world can be
represented as shown in FIG. 128B with the level as the horizontal
axis and the area in the image of the portion corresponding to that
level as the vertical axis, which shows that there is a relation in
area occupied in the image between the area corresponding to the
background in the image and the area of the portion corresponding
to the fine line.
In the same way, as shown in FIG. 129A, the image taken with the
sensor 2 is an image wherein there is a repeated array in the
direction in which the fine line exists of blocks which are made up
of pixels with the background level and the fine line level mixed
arrayed in the vertical direction in the pixel of the background
level, so that the mixture ratio thereof changes according to a
certain pattern, and accordingly, a mixed space region made up of
pixels occurring as the result of spatially mixing the background
and the fine line, of a level partway between the region which is
the background level (background region) and the fine line level,
as shown in FIG. 129B. Now, while the vertical axis in FIG. 129B is
the number of pixels, the area of one pixel is (d_CCD).sup.2, so it
can be said that the relation between the level of pixel values and
the number of pixels in FIG. 129B is the same as the relation
between the level of pixel values and distribution of area.
The same results are obtained regarding the portion enclosed with
the white line in the actual image shown in FIG. 130A (an image 31
pixels.times.31 pixels), as shown in FIG. 130B. As shown in FIG.
130B, the background portions shown in FIG. 130A (the portions
which appear black in FIG. 130A) has distribution of a great number
of pixels with low pixel value level (with pixel values around 20),
and these portions with little change make up the image of the
background region. Conversely, the portion wherein the pixel value
level in FIG. 130B is not low, i.e., pixels with pixel value level
distribution of around 40 to around 160 are pixels belonging to the
spatial mixture region which make up the image of the fine line,
and while the number of pixels for each pixel value is not great,
these are distributed over a wide range of pixel values.
Now, viewing the levels of each of the background and the fine line
in the real world image along the arrow direction (Y-coordinate
direction) shown in FIG. 131A for example, change occurs as shown
in FIG. 131B. That is to say, the background region from the start
of the arrow to the fine line has a relatively low background
level, and the fine line region has the fine line level which is a
high level, and passing the fine line region and returning to the
background region returns to the background level which is a low
level. As a result, this forms a pulse-shaped waveform where only
the fine line region is high level.
Conversely, in the image taken with the sensor 2, the relationship
between the pixel values of the pixels of the spatial direction
X=X1 in FIG. 132A corresponding to the arrow in FIG. 131A (the
pixels indicated by black dots in FIG. 132A) and the spatial
direction Y of these pixels is as shown in FIG. 132B. Note that in
FIG. 132A, between the two white lines extending toward the upper
right represents the fine line in the image of the real world.
That is to say, as shown in FIG. 132B, the pixel corresponding to
the center pixel in FIG. 132A has the highest pixel value, so the
pixel values of the pixels increases as the position of the spatial
direction Y moves from the lower part of the figure toward the
center pixel, and then gradually decreases after passing the center
position. As a result, as shown in FIG. 132B, peak-shaped waveforms
are formed. Also, the change in pixel values of the pixels
corresponding to the spatial directions X=X0 and X2 in FIG. 132A
also have the same shape, although the position of the peak in the
spatial direction Y is shifted according to the gradient of the
fine line.
Even in a case of an image actually taken with the sensor 2 as
shown in FIG. 133A for example, the same sort of results are
obtained, as shown in FIG. 133B. That is to say, FIG. 133B shows
the change in pixel values corresponding to the spatial direction Y
for each predetermined spatial direction X (in the figure, X=561,
562, 563) of the pixel values around fine line in the range
enclosed by the white lines in the image in FIG. 133A. In this way,
the image taken with the actual sensor 2 also has waveforms wherein
X=561 peaks at Y=730, X=562 at Y=705, and X=563 at Y=685.
Thus, while the waveform indicating change of level near the fine
line in the real world image exhibits a pulse-like waveform, the
waveform indicating change of pixel values in the image taken by
the sensor 2 exhibits peak-shaped waveforms.
That is to say, in other words, the level of the real world image
should be a waveform as shown in FIG. 131B, but distortion occurs
in the change in the imaged image due to having been taken by the
sensor 2, and accordingly it can be said that this has changed into
a waveform which is different from the real world image (wherein
information of the real world has dropped out), as shown in FIG.
132B.
Accordingly, a model (equivalent to the model 705 in FIG. 123) for
approximately describing the real world from the image data
obtained from the sensor 2 is set, in order to obtain continuity
information of the real world image from the image taken by the
sensor 2. For example, in the case of a fine line, the real world
image is set, as shown in FIG. 134. That is to say, parameters are
set with the level of the background portion at the left part of
the image as B1, the background portion at the right part of the
image as B2, the level of the fine line portion as L, the mixture
ratio of the fine line as .alpha., the width of the fine line as W,
and the angle of the fine line as to the horizontal direction as
.theta., this is formed into a model, a function approximately
expressing the real world is set, an approximation function which
approximately expresses the real world is obtained by obtaining the
parameters, and the direction (gradient or angle as to the
reference axis) of the fine line is obtained from the approximation
function.
At this time, the left part and right part of the background region
can be approximated as being the same, and accordingly are
integrated into B (=B1=B2) as shown in FIG. 135. Also, the width of
the fine line is to be one pixel or more. At the time of taking the
real world thus set with the sensor 2, the taken image is imaged as
shown in FIG. 136A. Note that in FIG. 136A, the space between the
two white lines extending towards the upper right represents the
fine line on the real world image.
That is to say, pixels existing in a position on the fine line of
the real world are of a level closest to the level of the fine
line, so the pixel value decreases the further away from the fine
line in the vertical direction (direction of the spatial direction
Y), and the pixel values of pixels which exist at positions which
do not come into contact with the fine line region, i.e.,
background region pixels, have pixel values of the background
value. At this time, the pixel values of the pixels existing at
positions straddling the fine line region and the background region
have pixel values wherein the pixel value B of the background level
and the pixel value L of the fine line level L are mixed with a
mixture ratio .alpha..
In the case of taking each of the pixels of the imaged image as the
pixel of interest in this way, the data acquiring unit 712 extracts
the pixels of an acquired block corresponding to the pixel of
interest, extracts a dynamic range block for each of the pixels
making up the extracted acquired block, and extracts from the
pixels making up the dynamic range block a pixel with a pixel value
which is the maximum value and a pixel with a pixel value which is
the minimum value. That is to say, as shown in FIG. 136A, in the
event of extracting pixels of a dynamic range block (e.g., the 7
pixels of pix1 through 7 surrounded by the black solid line in the
drawing) corresponding to a predetermined pixel in the acquired
block (the pixel pix4 regarding which a square is described with a
black solid line in one grid of the drawing), as shown in FIG.
136A, the image of the real world corresponding to each pixel is as
shown in FIG. 136B.
That is to say, as shown in FIG. 136B, with the pixel pix1, the
portion taking up generally 1/8 of the area to the left is the
background region, and the portion taking up generally 7/8 of the
area to the right is the fine line region. With the pixel pix2,
generally the entire region is the fine line region. With the pixel
pix3, the portion taking up generally 7/8 of the area to the left
is the fine line region, and the portion taking up generally 1/8 of
the area to the right is the background region. With the pixel
pix4, the portion taking up generally 2/3 of the area to the left
is the fine line region, and the portion taking up generally 1/3 of
the area to the right is the background region. With the pixel
pix5, the portion taking up generally 1/3 of the area to the left
is the fine line region, and the portion taking up generally 2/3 of
the area to the right is the background region. With the pixel
pix6, the portion taking up generally 1/8 of the area to the left
is the fine line region, and the portion taking up generally 7/8 of
the area to the right is the background region. Further, with the
pixel pix7, the entire region is the background region.
As a result, the pixel values of the pixels pix1 through 7 of the
dynamic range block shown in FIG. 136A and FIG. 136B are pixel
values wherein the background level and the fine line level are
mixed at a mixture ratio corresponding to the ratio of the fine
line region and the background region. That is to say, the mixture
ratio of background level: foreground level is generally 1:7 for
pixel pix1, generally 0:1 for pixel pix2, generally 1:7 for pixel
pix3, generally 1:2 for pixel pix4, generally 2:1 for pixel pix5,
generally 7:1 for pixel pix6, and generally 1:0 for pixel pix7.
Accordingly, of the pixel values of the pixels pix1 through 7 of
the dynamic range block that has been extracted, pixel pix2 is the
highest, followed by pixels pix1 and 3, and then in the order of
pixel value, pixels pix4, 5, 6, and 7. Accordingly, with the case
shown in FIG. 136B, the maximum value is the pixel value of the
pixel pix2, and the minimum value is the pixel value of the pixel
pix7.
Also, as shown in FIG. 137A, the direction of the fine line can be
said to be the direction in which pixels with maximum pixel values
continue, so the direction in which pixels with the maximum value
are arrayed is the direction of the fine line.
Now, the gradient G.sub.f1 indicating the direction of the fine
line is the ratio of change in the spatial direction Y (change in
distance) as to the unit distance in the spatial direction X, so in
the case of an illustration such as in FIG. 137A, the distance of
the spatial direction Y as to the distance of one pixel in the
spatial direction X in the drawing is the gradient G.sub.f1.
Change of pixel values in the spatial direction Y of the spatial
directions X0 through X2 is such that the peak waveform is repeated
at predetermined intervals for each spatial direction X, as shown
in FIG. 137B. As described above, the direction of the fine line is
the direction in which pixels with maximum value continue in the
image taken by the sensor 2, so the interval S in the spatial
direction Y where the maximum values in the spatial direction X are
is the gradient G.sub.f1 of the fine line. That is to say, as shown
in FIG. 137C, the amount of change in the vertical direction as to
the distance of one pixel in the horizontal direction is the
gradient G.sub.f1. Accordingly, with the horizontal direction
corresponding to the gradient thereof as the reference axis, and
the angle of the fine line thereto expressed as .theta., as shown
in FIG. 137C, the gradient G.sub.f1 (corresponding to the angle
with the horizontal direction as the reference axis) of the fine
line can be expressed in the relation shown in the following
Expression (69). .theta.=Tan.sup.-1(G.sub.f1)(=Tan.sup.-1(S))
(69)
Also, in the case of setting a model such as shown in FIG. 135, and
further assuming that the relationship between the pixel values of
the pixels in the spatial direction Y is such that the waveform of
the peaks shown in FIG. 137B is formed of perfect triangles (an
isosceles triangle waveform where the leading edge or trailing edge
change linearly), and, as shown in FIG. 138, with the maximum value
of pixel values of the pixels existing in the spatial direction Y,
in the spatial direction X of a predetermined pixel of interest as
Max=L (here, a pixel value corresponding to the level of the fine
line in the real world), and the minimum value as Min=B (here, a
pixel value corresponding to the level of the background in the
real world), the relationship illustrated in the following
Expression (70) holds. L-B=G.sub.f1.times.d.sub.--y (70)
Here, d_y indicates the difference in pixel values between pixels
in the spatial direction Y.
That is to say, the greater the gradient G.sub.f1 in the spatial
direction is, the closer the fine line is to being vertical, so the
waveform of the peaks is a waveform of isosceles triangles with a
great base, and conversely, the smaller the gradient S is, the
smaller the base of the isosceles triangles of the waveform is.
Consequently, the greater the gradient G.sub.f1 is, the smaller the
difference d_y of the pixel values between pixels in the spatial
direction Y is, and the smaller the gradient S is, the greater the
difference d_y of the pixel values between pixels in the spatial
direction Y is.
Accordingly, obtaining the gradient G.sub.f1 where the above
Expression (70) holds allows the angle .theta. of the fine line as
to the reference axis to be obtained. Expression (70) is a
single-variable function wherein G.sub.f1 is the variable, so this
could be obtained using one set of difference d_y of the pixel
values between pixels (in the vertical direction) around the pixel
of interest, and the difference between the maximum value and
minimum value (L-B), however, as described above, this uses an
approximation expression assuming that the change of pixel values
in the spatial direction Y assumes a perfect triangle, so dynamic
range blocks are extracted for each of the pixels of the extracted
block corresponding to the pixel of interest, and further the
dynamic range Dr is obtained from the maximum value and the minimum
value thereof, as well as statistically obtaining by the
least-square method, using the difference d_y of pixel values
between pixels in the spatial direction Y for each of the pixels in
the extracted block.
Now, before starting description of statistical processing by the
least-square method, first, the extracted block and dynamic range
block will be described in detail.
As shown in FIG. 139 for example, the extracted block may be three
pixels above and below the pixel of interest (the pixel of the grid
where a square is drawn with black solid lines in the drawing) in
the spatial direction Y, and one pixel to the right and left in the
spatial direction X, for a total of 15 pixels, or the like. Also,
in this case, for the difference d_y of pixel values between each
of the pixels in the extracted block, with difference corresponding
to pixel pix11 being expressed as d_y11 for example, in the case of
spatial direction X=X0, differences d_y11 through d_y16 are
obtained for the pixel values between the pixels pix11 and pix12,
pix12 and pix13, pix13 and pix14, pix15 and pix16, and pix16 and
pix17. At this time, the difference of pixel values between pixels
is obtained in the same way for spatial direction X=X1 and X2, as
well. As a result, there are 18 differences d_y of pixel values
between the pixels.
Further, with regard to the pixels of the extracted block,
determination has been made for this case based on the
determination results of the horizontal/vertical determining unit
711 that the pixels of the dynamic range block are, with regard to
pix11 for example, in the vertical direction, so as shown in FIG.
139, the pixel pix11 is taken along with three pixels in both the
upwards and downwards direction which is the vertical direction
(spatial direction Y) so that the range of the dynamic range block
B1 is 7 pixels, the maximum value and minimum value of the pixel
values of the pixels in this dynamic range block B1 is obtained,
and further, the dynamic range obtained from the maximum value and
the minimum value is taken as dynamic range Dr11. In the same way,
the dynamic range Dr12 is obtained regarding the pixel pix12 of the
extracted block from the 7 pixels of the dynamic range block B2
shown in FIG. 139 in the same way. Thus, the gradient G.sub.f1 is
statistically obtained using the least-square method, based on the
combination of the 18 pixel differences d_yi in the extracted block
and the corresponding dynamic ranges Dri.
Next, the single-variable least-square solution will be described.
Let us assume here that the determination results of the
horizontal/vertical determining unit 711 are the vertical
direction.
The single-variable least-square solution is for obtaining, for
example, the gradient G.sub.f1 of the straight line made up of
prediction values Dri_c wherein the distance to all of the actual
measurement values indicated by black dots in FIG. 140 is minimal.
Thus, the gradient S is obtained from the following technique based
on the relationship indicated in the above-described Expression
(70).
That is to say, with the difference between the maximum value and
the minimum value as the dynamic range Dr, the above Expression
(70) can be described as in the following Expression (71).
Dr=G.sub.f1.times.d.sub.--y (71)
Thus, the dynamic range Dri_c can be obtained by substituting the
difference d_yi between each of the pixels in the extracted block
into the above Expression (71). Accordingly, the relation of the
following Expression (72) is satisfied for each of the pixels.
Dri.sub.--c=G.sub.f1.times.d.sub.--yi (72)
Here, the difference d_yi is the difference in pixel values between
pixels in the spatial direction Y for each of the pixels i (for the
example, the difference in pixel values between pixels adjacent to
a pixel i in the upward direction or the downward direction, and
Dri_c is the dynamic range obtained when the Expression (70) holds
regarding the pixel i.
As described above, the least-square method as used here is a
method for obtaining the gradient G.sub.f1 wherein the sum of
squared differences Q of the dynamic range Dri_c for the pixel i of
the extracted block and the dynamic range Dri_r which is the actual
measured value of the pixel i, obtained with the method described
with reference to FIG. 136A and FIG. 136B, is the smallest for all
pixels within the image. Accordingly, the sum of squared
differences Q can be obtained by the following Expression (73).
.times..times..times..times..times..times. ##EQU00047##
The sum of squared differences Q shown in Expression (73) is a
quadratic function, which assumes a downward-convex curve as shown
in FIG. 141 regarding the variable G.sub.f1 (gradient G.sub.f1), so
G.sub.f1min where the gradient G.sub.f1 is the smallest is the
solution of the least-square method.
Differentiating the sum of squared differences Q shown in
Expression (73) with the variable G.sub.f1 yields dQ/dG.sub.f1
shown in the following Expression (74).
.differential..differential..times..times..times..times..times.
##EQU00048##
With Expression (74), 0 is the G.sub.f1 min assuming the minimal
value of the sum of squared differences Q shown in FIG. 141, so by
expanding the Expression wherein Expression (74) is 0 yields the
gradient G.sub.f1 with the following Expression (75).
.times..times..times..times. ##EQU00049##
The above Expression (75) is a so-called single-variable (gradient
G.sub.f1) normal equation.
Thus, substituting the obtained gradient G.sub.f1 into the above
Expression (69) yields the angle .theta. of the fine line with the
horizontal direction as the reference axis, corresponding to the
gradient G.sub.f1 of the fine line.
Now, in the above description, description has been made regarding
a case wherein the pixel of interest is a pixel on the fine line
which is within a range of angle .theta. of 45 degrees
degrees.ltoreq..theta.<135 degrees degrees with the horizontal
direction as the reference axis, but in the event that the pixel of
interest is a pixel on the fine line closer to the horizontal
direction, within a range of angle .theta. of 0 degrees
degrees.ltoreq..theta.<45 degrees degrees or 135 degrees
degrees.ltoreq..theta.<108 degrees degrees with the horizontal
direction as the reference axis for example, the difference of
pixel values between pixels adjacent to the pixel i in the
horizontal direction is d_xi, and in the same way, at the time of
obtaining the maximum value or minimum value of pixel values from
the multiple pixels corresponding to the pixel i, the pixels of the
dynamic range block to be extracted are selected from multiple
pixels existing in the horizontal direction as to the pixel i. With
the processing in this case, the relationship between the
horizontal direction and vertical direction in the above
description is simply switched, so description thereof will be
omitted.
Also, similar processing can be used to obtain the angle
corresponding to the gradient of a two-valued edge.
That is to say, enlarging the portion in an input image such as
that enclosed by the white lines as illustrated in FIG. 142A shows
that the edge portion in the image (the lower part of the
cross-shaped character written in white on a black banner in the
figure) (hereafter, an edge portion in an image made up of two
value levels will also be called a two-valued edge) is actually
displayed as shown in FIG. 142B. That is to say, in the real world,
the image has a boundary formed of the two types of levels of a
first level (the field level of the banner) and a second level (the
level of the character (the hatched portion with low concentration
in FIG. 142C)), and no other levels exist. Conversely, with the
image taken by the sensor 2, i.e., the image taken in increments of
pixels, a portion where first level pixels are arrayed and a
portion where second level pixels are arrayed border on a region
wherein there is a repeated array in the direction in which the
edge exists of blocks which are made up of pixels occurring as the
result of spatially mixing the first level and the second level,
arrayed in the vertical direction, so that the ratio (mixture
ratio) thereof changes according to a certain pattern.
That is to say, as shown in FIG. 143A, with regard to the spatial
direction X=X0, X1, and X2, the respective change of pixel values
in the spatial direction Y is such that as shown in FIG. 143B, the
pixel values are a predetermined minimum value pixel value from the
bottom of the figure to near to the two-valued edge (the straight
line in FIG. 143A which heads toward the upper right) boundary, but
the pixel value gradually increases near the two-valued edge
boundary, and at the point P.sub.E in the drawing past the edge the
pixel value reaches a predetermined maximum value. More
specifically, the change of the spatial direction X=X0 is such that
the pixel value gradually increases after passing the point P.sub.S
which is the minimum value of the pixel value, and reaches the
point P0 where the pixel value is the maximum value, as shown in
FIG. 143B. In comparison with this, the change of pixel values of
the pixels in the spatial direction X=X1 exhibits a waveform offset
in the spatial direction, and accordingly increases to the maximum
value of the pixel value via the point P1 in the drawing, with the
position where the pixel value gradually increases from the minimum
value of pixel values being a direction offset in the positive
direction of the spatial direction Y as shown in FIG. 143B.
Further, change of pixel values in the spatial direction Y at the
spatial direction X=X2 decreases via a point P2 in the drawing
which is even further shifted in the positive direction of the
spatial direction Y, and goes from the maximum value of the pixel
value to the minimum value.
A similar tendency can be observed at the portion enclosed with the
white line in the actual image, as well. That is to say, in the
portion enclosed with the white line in the actual image in FIG.
144A (a 31 pixel.times.31 pixel image), the background portion (the
portion which appears black in FIG. 144A) has distribution of a
great number of pixels with low pixel values (pixel value around
90) as shown in FIG. 144B, and these portions with little change
form the image of the background region. Conversely, the portion in
FIG. 144B wherein the pixel values are not low, i.e., pixels with
pixel values distributed around 100 to 200 are a distribution of
pixels belonging to the spatially mixed region between the
character region and the background region, and while the number of
pixels per pixel value is small, the distribution is over a wide
range of pixel values. Further, a great number of pixels of the
character region with high pixel values (the portion which appears
white in FIG. 144A) are distributed around the pixel value shown as
220.
As a result, the change of pixel values in the spatial direction Y
as to the predetermined spatial direction X in the edge image shown
in FIG. 145A is as shown in FIG. 145B.
That is, FIG. 145B illustrates the change of pixel values
corresponding to the spatial direction Y, for each predetermined
spatial direction X (in the drawing, X=658, 659, 660) regarding the
pixel values near the edge within the range enclosed by the white
lines in the image in FIG. 145A. As can be seen here, in the image
taken by the actual sensor 2 as well, with X=658, the pixel value
begins to increase around Y=374 (the distribution indicated by
black circles in the drawing), and reaches the maximum value around
X=382. Also, with X=659, the pixel value begins to increase around
Y=378 which is shifted in the positive direction as to the spatial
direction Y (the distribution indicated by black triangles in the
drawing), and reaches the maximum pixel value around X=386.
Further, with X=660, the pixel value begins to increase around
Y=382 which is shifted even further in the positive direction as to
the spatial direction Y (the distribution indicated by black
squares in the drawing), and reaches the maximum value around
X=390.
Accordingly, in order to obtain continuity information of the real
world image from the image taken by the sensor 2, a model is set to
approximately describe the real world from the image data acquired
by the sensor 2. For example, in the case of a two-valued edge, a
real world image is set, as shown in FIG. 146. That is to say,
parameters are set with the background portion level to the left in
the figure as V1, the character portion level to the right side in
the figure as V2, the mixture ratio between pixels around the
two-valued edge as .alpha., and the angle of the edge as to the
horizontal direction as .theta., this is formed into a model, a
function which approximately expresses the real world is set, the
parameters are obtained so as to obtain a function which
approximately expresses the real world, and the direction
(gradient, or angle as to the reference axis) of the edge is
obtained from the approximation function.
Now, the gradient indicating the direction of the edge is the ratio
of change in the spatial direction Y (change in distance) as to the
unit distance in the spatial direction X, so in a case such as
shown in FIG. 147A, the distance in the spatial direction Y as to
the distance of one pixel in the spatial direction X in the drawing
is the gradient.
The change in pixel values as to the spatial direction Y for each
of the spatial directions X0 through X2 is such that the same
waveforms are repeated at predetermined intervals for each of the
spatial directions X, as shown in FIG. 147B. As described above,
the edge in the image taken by the sensor 2 is the direction in
which similar pixel value change (in this case, change in pixel
values in a predetermined spatial direction Y, changing from the
minimum value to the maximum value) spatially continues, so the
intervals S of the position where change of pixel values in the
spatial direction Y begins, or the spatial direction Y which is the
position where change ends, for each of the spatial directions X,
is the gradient G.sub.fe of the edge. That is to say, as shown in
FIG. 147C, the amount of change in the vertical direction as to the
distance of one pixel in the horizontal direction, is the gradient
G.sub.fe.
Now, this relationship is the same as the relationship regarding
the gradient G.sub.f1 of the fine line described above with
reference to FIG. 137A through C. Accordingly, the relational
expression is the same. That is to say, the relational expression
in the case of a two-valued edge is that shown in FIG. 148, with
the pixel value of the background region as V1, and the pixel value
of the character region as V2, each as the minimum value and the
maximum value. Also, with the mixture ratio of pixels near the edge
as .alpha., and the edge gradient as G.sub.fe, relational
expressions which hold will be the same as the above Expression
(69) through Expression (71) (with G.sub.f1 replaced with
G.sub.fe).
Accordingly, the data continuity detecting unit 101 shown in FIG.
124 can detect the angle corresponding to the gradient of the fine
line, and the angle corresponding to the gradient of the edge, as
data continuity information with the same processing. Accordingly,
in the following, gradient will collectively refer to the gradient
of the fine line and the gradient of the two-valued edge, and will
be called gradient G.sub.f. Also, the gradient G.sub.f1 in the
above Expression (73) through Expression (75) may be G.sub.fe, and
consequently, will be considered to be substitutable with
G.sub.f.
Next, the processing for detecting data continuity will be
described with reference to the flowchart in FIG. 149.
In step S701, the horizontal/vertical determining unit 711
initializes a counter T which identifies each of the pixels of the
input image.
In step S702, the horizontal/vertical determining unit 711 performs
processing for extracting data necessary in later steps.
Now, the processing for extracting data will be described with
reference to the flowchart in FIG. 150.
In step S711, the horizontal/vertical determining unit 711 of the
data selecting unit 701 computes, for each pixel of interest T, as
described with reference to FIG. 125, the sum of difference
(activity) of the pixel values of the pixel values between the
pixels in the horizontal direction (hdiff) and the sum of
difference (activity) between pixels in the vertical direction
(vdiff), with regard to nine pixels adjacent in the horizontal,
vertical, and diagonal directions, and further obtains the
difference thereof the difference (hdiff minus vdiff); in the event
that (hdiff minus vdiff).gtoreq.0, and with the pixel of interest T
taking the horizontal direction as the reference axis,
determination is made that it is a pixel near a fine line or
two-valued edge closer to the vertical direction, wherein the angle
.theta. as to the reference axis is 45 degrees
degrees.ltoreq..theta.<135 degrees degrees, and determination
results indicating that the extracted block to be used corresponds
to the vertical direction are output to the data acquiring unit 712
and the data supplementing unit 702.
On the other hand, in the event that (hdiff minus vdiff)<0, and
with the pixel of interest taking the horizontal direction as the
reference axis, determination is made by the horizontal/vertical
determining unit 711 that it is a pixel near a fine line or edge
closer to the horizontal direction, wherein the angle .theta. of
the fine line or the two-valued edge as to the reference axis is 0
degrees degrees.ltoreq..theta.<45 degrees degrees or 135 degrees
degrees.ltoreq..theta.<180 degrees degrees, and determination
results indicating that the extracted block to be used corresponds
to the horizontal direction are output to the data acquiring unit
712 and the data supplementing unit 702.
That is, the gradient of the fine line or two-valued edge being
closer to the vertical direction means that, as shown in FIG. 131A
for example, the portion of the fine line which intersects with the
arrow in the drawing is greater, so extracted blocks with an
increased number of pixels in the vertical direction are set
(vertically long extracted blocks are set). In the same way, with
the case of fine lines having a gradient closer to the horizontal
direction, extracted blocks with an increased number of pixels in
the horizontal direction are set (horizontally long extracted
blocks are set). Thus, accurate maximum values and minimum values
can be computed without increasing the amount of unnecessary
calculations.
In step S712, the data acquiring unit 712 extracts pixels of an
extracted block corresponding to the determination results input
from the horizontal/vertical determining unit 711 indicating the
horizontal direction or the vertical direction for the pixel of
interest. That is to say, as shown in FIG. 139 for example, (three
pixels in the horizontal direction).times.(seven pixels in the
vertical direction) for a total of 21 pixels, centered on the pixel
of interest, are extracted as the extracted block, and stored.
In step S713, the data acquiring unit 712 extracts the pixels of
dynamic range blocks corresponding to the direction corresponding
to the determination results of the horizontal/vertical determining
unit 711 for each of the pixels in the extracted block, and stores
these. That is to say, as described above with reference to FIG.
139, in this case, with regard to the pixel pix11 of the extracted
block for example, the determination results of the
horizontal/vertical determining unit 711 indicate the vertical
direction, so the data acquiring unit 712 extracts the dynamic
range block B1 in the vertical direction, and extracts the dynamic
range block B2 for the pixel pix12 in the same way. Dynamic range
blocks are extracted for the other extracted blocks in the same
way.
That is to say, information of pixels necessary for computation of
the normal equation regarding a certain pixel of interest T is
stored in the data acquiring unit 712 with this data extracting
processing (a region to be processed is selected).
Now, let us return to the flowchart in FIG. 149.
In step S703, the data supplementing unit 702 performs processing
for supplementing the values necessary for each of the items in the
normal equation (Expression (74) here).
Now, the supplementing process to the normal equation will be
described with reference to the flowchart in FIG. 24.
In step S721, the difference supplementing unit 721 obtains
(detects) the difference of pixel values between the pixels of the
extracted block stored in the data acquiring unit 712, according to
the determination results of the horizontal/vertical determining
unit 711 of the data selecting unit 701, and further raises these
to the second power (squares) and supplements. That is to say, in
the event that the determination results of the horizontal/vertical
determining unit 711 are the vertical direction, the difference
supplementing unit 721 obtains the difference of pixel values
between pixels adjacent to each of the pixels of the extracted
block in the vertical direction, and further squares and
supplements these. In the same way, in the event that the
determination results of the horizontal/vertical determining unit
711 are the horizontal direction, the difference supplementing unit
721 obtains the difference of pixel values between pixels adjacent
to each of the pixels of the extracted block in the horizontal
direction, and further squares and supplements these. As a result,
the difference supplementing unit 721 generates the sum of squared
difference of the items to be the denominator in the
above-described Expression (75) and stores.
In step S722, the MaxMin acquiring unit 722 obtains the maximum
value and minimum value of the pixel values of the pixels contained
in the dynamic range block stored in the data acquiring unit 712,
and in step S723, obtains (detects) the dynamic range from the
maximum value and minimum value, and outputs this to the difference
supplementing unit 723. That is to say, in the case of a 7-pixel
dynamic range block made up of pixels pix1 through 7 as illustrated
in FIG. 136B, the pixel value of pix2 is detected as the maximum
value, the pixel value of pix7 is detected as the minimum value,
and the difference of these is obtained as the dynamic range.
In step S724, the difference supplementing unit 723 obtains
(detects), from the pixels in the extracted block stored in the
data acquiring unit 712, the difference in pixel values between
pixel adjacent in the direction corresponding to the determination
results of the horizontal/vertical determining unit 711 of the data
selecting unit 701, and supplements values multiplied by the
dynamic range input from the MaxMin acquiring unit 722. That is to
say, the difference supplementing unit 721 generates a sum of items
to serve as the numerator in the above-described Expression (75),
and stores this.
Now, let us return to description of the flowchart in FIG. 149.
In step S704, the difference supplementing unit 721 determines
whether or not the difference in pixel values between pixels (the
difference in pixel values between pixels adjacent in the direction
corresponding to the determination results of the
horizontal/vertical determining unit 711) has been supplemented for
all pixels of the extracted block, and in the event that
determination is made that, for example, the difference in pixel
values between pixels has not been supplemented for all pixels of
the extracted block, the flow returns to step S702, and the
subsequent processing is repeated. That is to say, the processing
of step S702 through S704 is repeated until determination is made
that the difference in pixel values between pixels has been
supplemented for all pixels of the extracted block.
In the event that determination is made in step S704 that the
difference in pixel values between pixels has been supplemented for
all pixels of the extracted block, in step S705, the difference
supplementing units 721 and 723 output the supplementing results
stored therein to the continuity direction derivation unit 703.
In step S706, the continuity direction computation unit 731 solves
the normal equation given in the above-described Expression (75),
based on: the sum of squared difference in pixel values between
pixels adjacent in the direction corresponding to the determination
results of the horizontal/vertical determining unit 711, of the
pixels in the acquired block input from the difference
supplementing unit 721 of the data supplementing unit 702; the
difference in pixel values between pixels adjacent in the direction
corresponding to the determination results of the
horizontal/vertical determining unit 711, of the pixels in the
acquired block input from the difference supplementing unit 723;
and the sum of products of the dynamic ranges corresponding to the
pixels of the obtained block; thereby statistically computing and
outputting the angle indicating the direction of continuity (the
angle indicating the gradient of the fine line or two-valued edge),
which is the data continuity information of the pixel of interest,
using the least-square method.
In step S707, the data acquiring unit 712 determines whether or not
processing has been performed for all pixels of the input image,
and in the event that determination is made that processing has not
been performed for all pixels of the input image for example, i.e.,
that information of the angle of the fine line or two-valued edge
has not been output for all pixels of the input image, the counter
T is incremented by 1 in step S708, and the process returns to step
S702. That is to say, the processing of steps S702 through S708 is
repeated until pixels of the input image to be processed are
changed and processing is performed for all pixels of the input
image. Change of pixel by the counter T may be according to raster
scan or the like for example, or may be sequential change according
to other rules.
In the event that determination is made in step S707 that
processing has been performed for all pixels of the input image, in
step S709 the data acquiring unit 712 determines whether or not
there is a next input image, and in the event that determination is
made that there is a next input image, the processing returns to
step S701, and the subsequent processing is repeated.
In the event that determination is made in step S709 that there is
no next input image, the processing ends.
According to the above processing, the angle of the fine line or
two-valued edge is detected as continuity information and
output.
The angle of the fine line or two-valued edge obtained by this
statistical processing approximately matches the angle of the fine
line or two-valued edge obtained using correlation. That is to say,
with regard to the image of the range enclosed by the white lines
in the image shown in FIG. 152A, as shown in FIG. 152B, the angle
indicating the gradient of the fine line obtained by the method
using correlation (the black circles in the figure) and the angle
of the fine line obtained by statistical processing with the data
continuity detecting unit 101 shown in FIG. 124 (the black
triangles in the figure) approximately agree at the spatial
direction Y coordinates near the fine line, with regard to change
in gradient in the spatial direction Y at predetermined coordinates
in the horizontal direction on the fine line. Note that in FIG.
152B, the spatial directions Y=680 through 730 between the black
lines in the figure are the coordinates on the fine line.
In the same way, with regard to the image of the range enclosed by
the white lines in the image shown in FIG. 153A, as shown in FIG.
153B, the angle indicating the gradient of the two-valued edge
obtained by the method using correlation (the black circles in the
figure) and the angle of the two-valued edge obtained by
statistical processing with the data continuity detecting unit 101
shown in FIG. 124 (the black triangles in the figure) approximately
agree at the spatial direction Y coordinates near the fine line,
with regard to change in gradient in the spatial direction Y at
predetermined coordinates in the horizontal direction on the
two-valued edge. Note that in FIG. 153B, the spatial directions
Y=(around) 376 through (around) 388 are the coordinates on the fine
line.
Consequently, the data continuity detecting unit 101 shown in FIG.
124 can statistically obtain the angle indicating the gradient of
the fine line or two-valued edge (the angle with the horizontal
direction as the reference axis here) using information around each
pixel for obtaining the angle of the fine line or two-valued edge
as the data continuity, unlike the method using correlation with
blocks made up of predetermined pixels, and accordingly, there is
no switching according to predetermined angle ranges as observed
with the method using correlation, thus, the angle of the gradients
of all fine lines or two-valued edges can be obtained with the same
processing, thereby enabling simplification of the processing.
Also, while description has been made above regarding an example of
the data continuity detecting unit 101 outputting the angle between
the fine line or two-valued edge and a predetermined reference axis
as the continuity information, but it is conceivable that depending
on the subsequent processing, outputting the angle as such may
improve processing efficiency. In such a case, the continuity
direction derivation unit 703 and continuity direction computation
unit 731 of the data continuity detecting unit 101 may output the
gradient G.sub.f of the fine line or two-valued edge obtained by
the least-square method as continuity information, without
change.
Further, while description has been made above regarding a case
wherein the dynamic range Dri_r in Expression (75) is computed
having been obtained regarding each of the pixels in the extracted
block, but setting the dynamic range block sufficiently great,
i.e., setting the dynamic range for a great number of pixels of
interest and a great number of pixels therearound, the maximum
value and minimum value of pixel values of pixels in the image
should be selected at all times for the dynamic range. Accordingly,
an arrangement may be made wherein computation is made for the
dynamic range Dri_r with the dynamic range Dri_r as a fixed value
obtained as the dynamic range from the maximum value and minimum
value of pixels in the extracted block or in the image data without
computing each pixel of the extracted block.
That is to say, an arrangement may be made to obtain the angle
.theta. (gradient G.sub.f) of the fine line by supplementing only
the difference in pixel values between the pixels, as in the
following Expression (76). Fixing the dynamic range in this way
allows the computation processing to be simplified, and processing
can be performed at high speed.
.times..times..times. ##EQU00050##
Next, description will be made regarding the data continuity
detecting unit 101 for detecting the mixture ratio of the pixels as
data continuity information with reference to FIG. 154.
Note that with the data continuity detecting unit 101 shown in FIG.
154, portions which correspond to those of the data continuity
detecting unit 101 shown in FIG. 124 are denoted with the same
symbols, and description thereof will be omitted as
appropriate.
With the data continuity detecting unit 101 shown in FIG. 154, what
differs from the data continuity detecting unit 101 shown in FIG.
124 is the point that a data supplementing unit 751 and mixture
ratio derivation unit 761 are provided instead of the data
supplementing unit 702 and continuity direction derivation unit
703.
A MaxMin acquiring unit 752 of the data supplementing unit 751
performs the same processing as the MaxMin acquiring unit 722 in
FIG. 124, and the maximum value and minimum value of the pixel
values of the pixels in the dynamic range block are obtained, the
difference (dynamic range) of the maximum value and minimum value
is obtained, and output to supplementing units 753 and 755 as well
as outputting the maximum value to a difference computing unit
754.
The supplementing unit 753 squares the value obtained by the MaxMin
acquiring unit, performs supplementing for all pixels of the
extracted block, obtains the sum thereof, and outputs to the
mixture ratio derivation unit 761.
The difference computing unit 754 obtains the difference between
each pixel in the acquired block of the data acquiring unit 712 and
the maximum value of the corresponding dynamic range block, and
outputs this to the supplementing unit 755.
The supplementing unit 755 multiplies the difference between the
maximum value and minimum value (dynamic range) of each pixel of
the acquired block input from the Max Min acquiring unit 752 with
the difference between the pixel value of each of the pixels in the
acquired block input from the difference computing unit 754 and the
maximum value of the corresponding dynamic range block, obtains the
sum thereof, and outputs to the mixture ratio derivation unit
761.
A mixture ratio calculating unit 762 of the mixture ratio
derivation unit 761 statistically obtains the mixture ratio of the
pixel of interest by the least-square method, based on the values
input from the supplementing units 753 and 755 of the data
supplementing unit, and outputs this as data continuity
information.
Next, the mixture ratio derivation method will be described.
As shown in FIG. 155A, in the event that a fine line exists on the
image, the image taken with the sensor 2 is an image such as shown
in FIG. 155B. In this image, let us hold in interest the pixel
enclosed by the black solid lines on the spatial direction X=X1 in
FIG. 155B. Note that the range between the white lines in FIG. 155B
indicates the position corresponding to the fine line region in the
real world. The pixel value M of this pixel should be an
intermediate color between the pixel value B corresponding to the
level of the background region, and the pixel value L corresponding
to the level of the fine line region, and in further detail, this
pixel value P.sub.S should be a mixture of each level according to
the area ratio between the background region and fine line region.
Accordingly, the pixel value P.sub.S can be expressed by the
following Expression (77).
P.sub.s=.alpha..times.B+(1-.alpha.).times.L (77)
Here, .alpha. is the mixture ratio, and more specifically,
indicates the ratio of area which the background region occupies in
the pixel of interest. Accordingly, (1-.alpha.) can be said to
indicate the ratio of area which the fine line region occupies.
Now, pixels of the background region can be considered to be the
component of an object existing in the background, and thus can be
said to be a background object component. Also, pixels of the fine
line region can be considered to be the component of an object
existing in the foreground as to the background object, and thus
can be said to be a foreground object component.
Consequently, the mixture ratio .alpha. can be expressed by the
following Expression (78) by expanding the Expression (77).
.alpha.=(P.sub.s-L)/(B-L) (78)
Further, in this case, we are assuming that the pixel value exists
at a position straddling the first pixel value (pixel value B)
region and the second pixel value (pixel value L) region, and
accordingly, the pixel value L can be substituted with the maximum
value Max of the pixel values, and further, the pixel value B can
be substituted with the minimum value of the pixel value.
Accordingly, the mixture ratio .alpha. can also be expressed by the
following Expression (79). .alpha.=(P.sub.s-Max)/(Min-Max) (79)
As a result of the above, the mixture ratio .alpha. can be obtained
from the dynamic range (equivalent to (Min-Max)) of the dynamic
range block regarding the pixel of interest, and the difference
between the pixel of interest and the maximum value of pixels
within the dynamic range block, but in order to further improve
precision, the mixture ratio .alpha. will here be statistically
obtained by the least-square method.
That is to say, expanding the above Expression (79) yields the
following Expression (80). (P.sub.S-Max)=.alpha..times.(Min-Max)
(80)
As with the case of the above-described Expression (71), this
Expression (80) is a single-variable least-square equation. That is
to say, in Expression (71), the gradient G.sub.f was obtained by
the least-square method, but here, the mixture ratio .alpha. is
obtained. Accordingly, the mixture ratio .alpha. can be
statistically obtained by solving the normal equation shown in the
following Expression (81).
.alpha..times..times..times..times. ##EQU00051##
Here, i is for identifying the pixels of the extracted block.
Accordingly, in Expression (81), the number of pixels in the
extracted block is n.
Next, the processing for detecting data continuity with the mixture
ratio as data continuity will be described with reference to the
flowchart in FIG. 156.
In step S731, the horizontal/vertical determining unit 711
initializes the counter U which identifies the pixels of the input
image.
In step S732, the horizontal/vertical determining unit 711 performs
processing for extracting data necessary for subsequent processing.
Note that the processing of step S732 is the same as the processing
described with reference to the flowchart in FIG. 150, so
description thereof will be omitted.
In step S733, the data supplementing unit 751 performs processing
for supplementing values necessary of each of the items for
computing the normal equation (Expression (81) here).
Now, the processing for supplementing to the normal equation will
be described with reference to the flowchart in FIG. 157.
In step S751, the MaxMin acquiring unit 752 obtains the maximum
value and minimum value of the pixels values of the pixels
contained in the dynamic range block stored in the data acquiring
unit 712, and of these, outputs the minimum value to the difference
computing unit 754.
In step S752, the MaxMin acquiring unit 752 obtains the dynamic
range from the difference between the maximum value and the minimum
value, and outputs this to the difference supplementing units 753
and 755.
In step S753, the supplementing unit 753 squares the dynamic range
(Max-Min) input from the MaxMin acquiring unit 752, and
supplements. That is to say, the supplementing unit 753 generates
by supplementing a value equivalent to the denominator in the above
Expression (81).
In step S754, the difference computing unit 754 obtains the
difference between the maximum value of the dynamic range block
input from the MaxMin acquiring unit 752 and the pixel values of
the pixels currently being processed in the extracted block, and
outputs to the supplementing unit 755.
In step S755, the supplementing unit 755 multiplies the dynamic
range input from the MaxMin acquiring unit 752 with the difference
between the pixel values of the pixels currently being processed
input from the difference computing unit 754 and the maximum value
of the pixels of the dynamic range block, and supplements. That is
to say, the supplementing unit 755 generates values equivalent to
the numerator item of the above Expression (81).
As described above, the data supplementing unit 751 performs
computation of the items of the above Expression (81) by
supplementing.
Now, let us return to the description of the flowchart in FIG.
156.
In step S734, the difference supplementing unit 721 determines
whether or not supplementing has ended for all pixels of the
extracted block, and in the event that determination is made that
supplementing has not ended for all pixels of the extracted block
for example, the processing returns to step S732, and the
subsequent processing is repeated. That is to say, the processing
of steps S732 through S734 is repeated until determination is made
that supplementing has ended for all pixels of the extracted
block.
In step S734, in the event that determination is made that
supplementing has ended for all pixels of the extracted block, in
step S735 the supplementing units 753 and 755 output the
supplementing results stored therein to the mixture ratio
derivation unit 761.
In step S736, the mixture ratio calculating unit 762 of the mixture
ratio derivation unit 761 statistically computes, by the
least-square method, and outputs, the mixture ratio of the pixel of
interest which is the data continuity information, by solving the
normal equation shown in Expression (81), based on the sum of
squares of the dynamic range, and the sum of multiplying the
difference between the pixel values of the pixels of the extracted
block and the maximum value of the dynamic block by the dynamic
range, input from the supplementing units 753 and 755 of the data
supplementing unit 751.
In step S737, the data acquiring unit 712 determines whether or not
processing has been performed for all pixels in the input image,
and in the event that determination is made that, for example,
processing has not been performed for all pixels in the input
image, i.e., in the event that determination is made that the
mixture ratio has not been output for all pixels of the input
image, in step S738 the counter U is incremented by 1, and the
processing returns to step S732.
That is to say, the processing of steps S732 through S738 is
repeated until pixels to be processed within the input image are
changed and the mixture ratio is computed for all pixels of the
input image. Change of pixel by the counter U may be according to
raster scan or the like for example, or may be sequential change
according to other rules.
In the event that determination is made in step S737 that
processing has been performed for all pixels of the input image, in
step S739 the data acquiring unit 712 determines whether or not
there is a next input image, and in the event that determination is
made that there is a next input image, the processing returns to
step S731, and the subsequent processing is repeated.
In the event that determination is made in step S739 that there is
no next input image, the processing ends.
Due to the above processing, the mixture ratio of the pixels is
detected as continuity information, and output.
FIG. 158B illustrates the change in the mixture ratio on
predetermined spatial directions X (=561, 562, 563) with regard to
the fine line image within the white lines in the image shown in
FIG. 158A, according to the above technique, for example. As shown
in FIG. 158B, the change in the mixture ratio in the spatial
direction Y which is continuous in the horizontal direction is such
that, respectively, in the case of the spatial direction X=563, the
mixture ratio starts rising at around the spatial direction Y=660,
peaks at around Y=685, and drops to Y=710. Also, in the case of the
spatial direction X=562, the mixture ratio starts rising at around
the spatial direction Y=680, peaks at around Y=705, and drops to
Y=735. Further, in the case of the spatial direction X=561, the
mixture ratio starts rising at around the spatial direction Y=705,
peaks at around Y=725, and drops to Y=755.
Thus, as shown in FIG. 158B, the change of each of the mixture
ratios in the continuous spatial directions X is the same change as
the change in pixel values changing according to the mixture ratio
(the change in pixel values shown in FIG. 133B), and is cyclically
continuous, so it can be understood that the mixture ratio of
pixels near the fine line are being accurately represented.
Also, in the same way, FIG. 159B illustrates the change in the
mixture ratio on predetermined spatial directions X (=658, 659,
660) with regard to the two-valued edge image within the white
lines in the image shown in FIG. 159A. As shown in FIG. 159B, the
change in the mixture ratio in the spatial direction Y which is
continuous in the horizontal direction is such that, respectively,
in the case of the spatial direction X=660, the mixture ratio
starts rising at around the spatial direction Y=750, and peaks at
around Y=765. Also, in the case of the spatial direction X=659, the
mixture ratio starts rising at around the spatial direction Y=760,
and peaks at around Y=775. Further, in the case of the spatial
direction X=658, the mixture ratio starts rising at around the
spatial direction Y=770, and peaks at around Y=785.
Thus, as shown in FIG. 159B, the change of each of the mixture
ratios of the two-valued edge is approximately the same as change
which is the same change as the change in pixel values changing
according to the mixture ratio (the change in pixel values shown in
FIG. 145B), and is cyclically continuous, so it can be understood
that the mixture ratio of pixel values near the two-valued edge are
being accurately represented.
According to the above, the mixture ratio of each pixel can be
statistically obtained as data continuity information by the
least-square method. Further, the pixel values of each of the
pixels can be directly generated based on this mixture ratio.
Also, if we say that the change in mixture ratio has continuity,
and further, the change in the mixture ratio is linear, the
relationship such as indicated in the following Expression (82)
holds. .alpha.=m.times.y+n (82)
Here, m represents the gradient when the mixture ratio .alpha.
changes as to the spatial direction Y, and also, n is equivalent to
the intercept when the mixture ratio .alpha. changes linearly.
That is, as shown in FIG. 160, the straight line indicating the
mixture ratio is a straight line indicating the boundary between
the pixel value B equivalent to the background region level and the
level L equivalent to the fine line level, and in this case, the
amount in change of the mixture ratio upon progressing a unit
distance with regard to the spatial direction Y is the gradient
m.
Accordingly, substituting Expression (82) into Expression (77)
yields the following Expression (83).
M=(m.times.y+n).times.B+(1-(m.times.y+n)).times.L (83)
Further, expanding this Expression (83) yields the following
Expression (84). M-L=(y.times.B-y.times.L).times.m+(B-L).times.n
(84)
In Expression (84), the first item m represents the gradient of the
mixture ratio in the spatial direction, and the second item is the
item representing the intercept of the mixture ratio. Accordingly,
an arrangement may be made wherein a normal equation is generated
using the least-square of two variables to obtain m and n in
Expression (84) described above.
However, the gradient m of the mixture ratio .alpha. is the
above-described gradient of the fine line or two-valued edge (the
above-described gradient G.sub.f) itself, so an arrangement may be
made wherein the above-described method is used to obtain the
gradient G.sub.f of the fine line or two-valued edge beforehand,
following which the gradient is used and substituted into
Expression (84), thereby making for a single-variable function with
regard to the item of the intercept, and obtaining with the
single-variable least-square method the same as the technique
described above.
While the above example has been described regarding a data
continuity detecting unit 101 for detecting the angle (gradient) or
mixture ratio of a fine line or two-valued edge in the spatial
direction as data continuity information, an arrangement may be
made wherein that corresponding to the angle in the spatial
direction obtained by replacing one of the spatial-direction axes
(spatial directions X and Y), for example, with the time-direction
(frame direction) T axis. That is to say, that which corresponds to
the angle obtained by replacing one of the spatial-direction axes
(spatial directions X and Y) with the time-direction (frame
direction) T axis, is a vector of movement of an object (movement
vector direction).
More specifically, as shown in FIG. 161A, in the event that an
object is moving upwards in the drawing with regard to the spatial
direction Y over time, the track of movement of the object is
manifested at the portion equivalent to the fine line in the
drawing (in comparison with that in FIG. 131A). Accordingly, the
gradient at the fine line in the time direction T represents the
direction of movement of the object (angle indicating the movement
of the object) (is equivalent to the direction of the movement
vector) in FIG. 161A. Accordingly, in the real world, in a frame of
a predetermined point-in-time indicated by the arrow in FIG. 161A,
a pulse-shaped waveform wherein the portion to be the track of the
object is the level of (the color of) the object, and other
portions are the background level, as shown in FIG. 161B, is
obtained.
In this way, in the case of imaging an object with movement with
the sensor 2, as shown in FIG. 162A, the distribution of pixel
values of each of the pixels of the frames from point-in-time T1
through T3 each assumes a peak-shaped waveform in the spatial
direction Y, as shown in FIG. 162B. This relationship can be
thought to be the same as the relationship in the spatial
directions X and Y, described with reference to FIG. 132A and FIG.
132B. Accordingly, in the event that the object has movement in the
frame direction T, the direction of the movement vector of the
object can be obtained as data continuity information in the same
way as with the information of the gradient of the fine line or the
angle (gradient) of the two-valued edge described above. Note that
in FIG. 162B, each grid in the frame direction T (time direction T)
is the shutter time making up the image of one frame.
Also, in the same way, in the event that there is movement of an
object in the spatial direction Y for each frame direction T as
shown in FIG. 163A, each pixel value corresponding to the movement
of the object as to the spatial direction Y on a frame
corresponding to a predetermined point-in-time T1 can be obtained
as shown in FIG. 163B. At this time, the pixel value of the pixel
enclosed by the black solid lines in FIG. 163B is a pixel value
wherein the background level and the object level are mixed in the
frame direction at a mixture ratio .beta., corresponding to the
movement of the object, as shown in FIG. 163C, for example.
This relationship is the same as the relationship described with
reference to FIG. 155A, FIG. 155B, and FIG. 155C.
Further, as shown in FIG. 164, the level 0 of the object and the
level B of the background can also be made to be linearly
approximated by the mixture ratio .beta. in the frame direction
(time direction). This relationship is the same relationship as the
linear approximation of mixture ratio in the spatial direction,
described with reference to FIG. 160.
Accordingly, the mixture ratio .beta. in the time (frame) direction
can be obtained as data continuity information with the same
technique as the case of the mixture ratio .alpha. in the spatial
direction.
Also, an arrangement may be made wherein the frame direction, or
one dimension of the spatial direction, is selected, and the data
continuity angle or the movement vector direction is obtained, and
in the same way, the mixture ratios .alpha. and .beta. may be
selectively obtained.
According to the above, light signals of the real world are
projected, a region, corresponding to a pixel of interest in the
image data of which a part of the continuity of the real world
light signals has dropped out, is selected, features for detecting
the angle as to a reference axis of the image data continuity
corresponding to the lost real world light signal continuity are
detected in the selected region, the angle is statistically
detected based on the detected features, and light signals are
estimated by estimating the lost real world light signal continuity
based on the detected angle of the continuity of the image data as
to the reference axis, so the angle of continuity (direction of
movement vector) or (a time-space) mixture ratio can be
obtained.
Next, description will be made, with reference to FIG. 165, of a
data continuity information detecting unit 101 which outputs, as
data continuity information, information of regions where
processing using data continuity information should be
performed.
An angle detecting unit 801 detects, of the input image, the
spatial-direction angle of regions having continuity, i.e., of
portions configuring fine lines and two-valued edges having
continuity in the image, and outputs the detected angle to an
actual world estimating unit 802. Note that this angle detecting
unit 801 is the same as the data continuity detecting unit 101 in
FIG. 3.
The actual world estimating unit 802 estimates the actual world
based on the angle indicating the direction of data continuity
input from the angle detecting unit 801, and information of the
input image. That is to say, the actual world estimating unit 802
obtains a coefficient of an approximation function which
approximately describes the intensity distribution of the actual
world light signals, from the input angle and each pixel of the
input image, and outputs to an error computing unit 803 the
obtained coefficient as estimation results of the actual world.
Note that this actual world estimating unit 802 is the same as the
actual world estimating unit 102 shown in FIG. 3.
The error computing unit 803 formulates an approximation function
indicating the approximately described real world light intensity
distribution, based on the coefficient input from the actual world
estimating unit 802, and further, integrates the light intensity
corresponding to each pixel position based on this approximation
function, thereby generating pixel values of each of the pixels
from the light intensity distribution estimated from the
approximation function, and outputs to a comparing unit 804 with
the difference as to the actually-input pixel values as error.
The comparing unit 804 compares the error input from the error
computing unit 803 for each pixel, and a threshold value set
beforehand, so as to distinguish between processing regions where
pixels exist regarding which processing using continuity
information is to be performed, and non-processing regions, and
outputs region information, distinguishing between processing
regions where processing using continuity information is to be
performed and non-processing regions, as continuity
information.
Next, description will be made regarding continuity detection
processing using the data continuity detecting unit 101 in FIG. 165
with reference to the flowchart in FIG. 166.
The angle detecting unit 801 acquires an image input in step S801,
and detects an angle indicating the direction of continuity in step
S802. More particularly, the angle detecting unit 801 detects a
fine line when the horizontal direction is taken as a reference
axis, or an angle indicating the direction of continuity having a
two-valued edge for example, and outputs this to the actual world
estimating unit 802.
In step S803, the actual world estimating unit 802 obtains a
coefficient of an approximation function f(x) made up of a
polynomial, which approximately describes a function F(x)
expressing the real world, based on angular information input from
the angle detecting unit 801 and input image information, and
outputs this to the error calculation unit 803. That is to say, the
approximation function f(x) expressing the real world is shown with
a primary polynomial such as the following Expression (85).
.function..times..times..times..times..times. ##EQU00052##
Here, wi is a coefficient of the polynomial, and the actual world
estimating unit 802 obtains this coefficient wi and outputs this to
the error calculation unit 803. Further, a gradient from the
direction of continuity can be obtained based on an angle input
from the angle detecting unit 801 (G.sub.f=tan.sup.-1.theta.,
G.sub.f: gradient, .theta.: angle), so the above Expression (85)
can be described with a quadratic polynomial such as shown in the
following Expression (86) by substituting a constraint condition of
this gradient G.sub.f.
.function..times..function..alpha..times..times..alpha..times..times..tim-
es..times..function..alpha..times..times..times..times..function..alpha..t-
imes..times. ##EQU00053##
That is to say, the above Expression (86) describes a quadratic
function f(x, y) obtained by expressing the width of a shift
occurring due to the primary approximation function f(x) described
with Expression (85) moving in parallel with the spatial direction
Y using a shift amount .alpha. (=-dy/G.sub.f: dy is the amount of
change in the spatial direction Y).
Accordingly, the actual world estimating unit 802 solves each
coefficient wi of the above Expression (86) using an input image
and angular information in the direction of continuity, and outputs
the obtained coefficients wi to the error calculation unit 803.
Here, description will return to the flowchart in FIG. 166.
In step S804, the error calculation unit 803 performs reintegration
regarding each pixel based on the coefficients input by the actual
world estimating unit 802. More specifically, the error calculation
unit 803 subjects the above Expression (86) to integration
regarding each pixel such as shown in the following Expression (87)
based on the coefficients input from the actual world estimating
unit 802.
.times..intg..times..intg..times..function..times.d.times.d.times..intg..-
times..intg..times..times..function..alpha..times..times..times.d.times.d.-
times..times..times..intg..times..intg..times..alpha..times..times..times.-
d.times.d.times..times..times..times..times..times..alpha..times..times..a-
lpha..function..alpha..function..times..alpha..times..times..alpha..times.-
.times. ##EQU00054##
Here, S.sub.S denotes the integrated result in the spatial
direction shown in FIG. 167. Also, the integral range thereof is,
as shown in FIG. 167, x.sub.m through X.sub.m+B for the spatial
direction X, and y.sub.m through y.sub.m+A for the spatial
direction Y. Also, in FIG. 167, let us say that each grid (square)
denotes one pixel, and both grid for the spatial direction X and
grid for the spatial direction Y is 1.
Accordingly, the error calculation unit 803, as shown in FIG. 168,
subjects each pixel to an integral arithmetic operation such as
shown in the following Expression (88) with an integral range of
x.sub.m through x.sub.m+1 for the spatial direction X of a curved
surface shown in the approximation function f(x, y), and y.sub.m
through y.sub.m+1 for the spatial direction Y (A=B=1), and
calculates the pixel value P.sub.S of each pixel obtained by
spatially integrating the approximation function expressing the
actual world in an approximate manner.
.times..intg..times..intg..times..function..times.d.times.d.times..intg..-
times..intg..times..times..function..alpha..times..times..times..times.d.t-
imes.d.times..times..times..intg..times..intg..times..alpha..times..times.-
.times.d.times.d.times..times..times..times..times..times..alpha..times..t-
imes..alpha..function..alpha..function..times..alpha..times..times..alpha.-
.times..times. ##EQU00055##
In other words, according to this processing, the error calculation
unit 803 serves as, so to speak, a kind of pixel value generating
unit, and generates pixel values from the approximation
function.
In step S805, the error calculation unit 803 calculates the
difference between a pixel value obtained with integration such as
shown in the above Expression (88) and a pixel value of the input
image, and outputs this to the comparison unit 804 as an error. In
other words, the error calculation unit 803 obtains the difference
between the pixel value of a pixel corresponding to the integral
range (x.sub.m through x.sub.m+1 for the spatial direction X, and
y.sub.m through y.sub.m+1 for the spatial direction Y) shown in the
above FIG. 167 and FIG. 168, and a pixel value obtained with the
integrated result in a range corresponding to the pixel as an
error, and outputs this to the comparison unit 804.
In step S806, the comparison unit 804 determines regarding whether
or not the absolute value of the error between the pixel value
obtained with integration input from the error calculation unit 803
and the pixel value of the input image is a predetermined threshold
value or less.
In step S806, in the event that determination is made that the
error is the predetermined threshold value or less, since the pixel
value obtained with integration is a value close to the pixel value
of the pixel of the input image, the comparison unit 804 regards
the approximation function set for calculating the pixel value of
the pixel as a function sufficiently approximated with the light
intensity allocation of a light signal in the real world, and
recognizes the region of the pixel now processed as a processing
region where processing using the approximation function based on
continuity information is performed in step S807. In further
detail, the comparison unit 804 stores the pixel now processed in
unshown memory as the pixel in the subsequent processing
regions.
On the other hand, in the event that determination is made that the
error is not the threshold value or less in step S806, since the
pixel value obtained with integration is a value far from the
actual pixel value, the comparison unit 804 regards the
approximation function set for calculating the pixel value of the
pixel as a function insufficiently approximated with the light
intensity allocation of a light signal in the real world, and
recognizes the region of the pixel now processed as a
non-processing region where processing using the approximation
function based on continuity information is not performed at a
subsequent stage in step S808. In further detail, the comparison
unit 804 stores the region of the pixel now processed in unshown
memory as the subsequent non-processing regions.
In step S809, the comparison unit 804 determines regarding whether
or not the processing has been performed as to all of the pixels,
and in the event that determination is made that the processing has
not been performed as to all of the pixels, the processing returns
to step S802, wherein the subsequent processing is repeatedly
performed. In other words, the processing in steps S802 through
S809 is repeatedly performed until determination processing wherein
comparison between a pixel value obtained with integration and a
pixel value input is performed, and determination is made regarding
whether or not the pixel is a processing region, is completed
regarding all of the pixels.
In step S809, in the event that determination is made that
determination processing wherein comparison between a pixel value
obtained with reintegration and a pixel value input is performed,
and determination is made regarding whether or not the pixel is a
processing region, has been completed regarding all of the pixels,
the comparison unit 804, in step S810, outputs region information
wherein a processing region where processing based on the
continuity information in the spatial direction is performed at
subsequent processing, and a non-processing region where processing
based on the continuity information in the spatial direction is not
performed are identified regarding the input image stored in the
unshown memory, as continuity information.
According to the above processing, based on the error between the
pixel value obtained by the integrated result in a region
corresponding to each pixel using the approximation function f(x)
calculated based on the continuity information and the pixel value
in the actual input image, evaluation for reliability of expression
of the approximation function is performed for each region (for
each pixel), and accordingly, a region having a small error, i.e.,
only a region where a pixel of which the pixel value obtained with
integration based on the approximation function is reliable exists
is regarded as a processing region, and the regions other than this
region are regarded as non-processing regions, and consequently,
only a reliable region can be subjected to the processing based on
the continuity information in the spatial direction, and the
necessary processing alone can be performed, whereby processing
speed can be improved, and also the processing can be performed as
to the reliable region alone, resulting in preventing image quality
due to this processing from deterioration.
Next, description will be made regarding other embodiments
regarding the data continuity information detecting unit 101 which
outputs region information where a pixel to be processed using data
continuity information exists, as data continuity information with
reference to FIG. 169.
A movement detecting unit 821 detects, of images input, a region
having continuity, i.e., movement having continuity in the frame
direction on an image (direction of movement vector: V.sub.f), and
outputs the detected movement to the actual world estimating unit
822. Note that this movement detecting unit 821 is the same as the
data continuity detecting unit 101 in FIG. 3.
The actual world estimating unit 822 estimates the actual world
based on the movement of the data continuity input from the
movement detecting unit 821, and the input image information. More
specifically, the actual world estimating unit 822 obtains
coefficients of the approximation function approximately describing
the intensity allocation of a light signal in the actual world in
the frame direction (time direction) based on the movement input
and each pixel of the input image, and outputs the obtained
coefficients to the error calculation unit 823 as an estimated
result in the actual world. Note that this actual world estimating
unit 822 is the same as the actual world estimating unit 102 in
FIG. 3.
The error calculation unit 823 makes up an approximation function
indicating the intensity allocation of light in the real world in
the frame direction, which is approximately described based on the
coefficients input from the actual world estimating unit 822,
further integrates the intensity of light equivalent to each pixel
position for each frame from this approximation function, generates
the pixel value of each pixel from the intensity allocation of
light estimated by the approximation function, and outputs the
difference with the pixel value actually input to the comparison
unit 824 as an error.
The comparison unit 824 identifies a processing region where a
pixel to be subjected to processing using the continuity
information exists, and a non-processing region by comparing the
error input from the error calculation unit 823 regarding each
pixel with a predetermined threshold value set beforehand, and
outputs region information wherein a processing region where
processing is performed using this continuity information and a
non-processing region are identified, as continuity
information.
Next, description will be made regarding continuity detection
processing using the data continuity detecting unit 101 in FIG. 169
with reference to the flowchart in FIG. 170.
The movement detecting unit 801 acquires an image input in step
S821, and detects movement indicating continuity in step S822. In
further detail, the movement detecting unit 801 detects movement of
a substance moving within the input image (direction of movement
vector: V.sub.f) for example, and outputs this to the actual world
estimating unit 822.
In step S823, the actual world estimating unit 822 obtains
coefficients of a function f(t) made up of a polynomial, which
approximately describes a function F(t) in the frame direction,
which expresses the real world, based on the movement information
input from the movement detecting unit 821 and the information of
the input image, and outputs this to the error calculation unit
823. That is to say, the function f(t) expressing the real world is
shown as a primary polynomial such as the following Expression
(89).
.function..times..times..times..times..times. ##EQU00056##
Here, wi is coefficients of the polynomial, and the actual world
estimating unit 822 obtains these coefficients wi, and outputs
these to the error calculation unit 823. Further, movement as
continuity can be obtained by the movement input from the movement
detecting unit 821 (V.sub.f=tan.sup.-1.theta.v, V.sub.f: gradient
in the frame direction of a movement vector, .theta.v: angle in the
frame direction of a movement vector), so the above Expression (89)
can be described with a quadratic polynomial such as shown in the
following Expression (90) by substituting a constraint condition of
this gradient.
.function..times..function..alpha..times..times..function..alpha..times..-
times..function..alpha..times..times..times..times..function..alpha..times-
..times. ##EQU00057##
That is to say, the above Expression (90) describes a quadratic
function f(t, y) obtained by expressing the width of a shift
occurring by a primary approximation function f(t), which is
described with Expression (89), moving in parallel to the spatial
direction Y, as a shift amount .alpha.t (=-dy/V.sub.f: dy is the
amount of change in the spatial direction Y).
Accordingly, the actual world estimating unit 822 solves each
coefficient wi of the above Expression (90) using the input image
and continuity movement information, and outputs the obtained
coefficients wi to the error calculation unit 823.
Now, description will return to the flowchart in FIG. 170.
In step S824, the error calculation unit 823 performs integration
regarding each pixel in the frame direction from the coefficients
input by the actual world estimating unit 822. That is to say, the
error calculation unit 823 integrates the above Expression (90)
regarding each pixel from coefficients input by the actual world
estimating unit 822 such as shown in the following Expression
(91).
.times..intg..times..intg..times..function..times.d.times.d.times..intg..-
times..intg..times..times..function..alpha..times..times..times.d.times.d.-
times..times..times..intg..times..intg..times..alpha..times..times..times.-
d.times.d.times..times..times..times..times..times..alpha..times..times..a-
lpha..function..alpha..function..times..alpha..times..times..alpha..times.-
.times. ##EQU00058##
Here, S.sub.t represents the integrated result in the frame
direction shown in FIG. 171. The integral range thereof is, as
shown in FIG. 171, T.sub.m through T.sub.m+B for the frame
direction T, and y.sub.m through y.sub.m+A for the spatial
direction Y. Also, in FIG. 171, let us say that each grid (square)
denotes one pixel, and both for the frame direction T and spatial
direction Y are 1. Here, "1 regarding the frame direction T" means
that the shutter time for the worth of one frame is 1.
Accordingly, the error calculation unit 823 performs, as shown in
FIG. 172, an integral arithmetic operation such as shown in the
following Expression (92) regarding each pixel with an integral
range of T.sub.m through T.sub.m+1 for the spatial direction T of a
curved surface shown in the approximation function f (t, y), and
y.sub.m through y.sub.m+1 for the spatial direction Y (A=B=1), and
calculates the pixel value P.sub.t of each pixel obtained from the
function approximately expressing the actual world.
.times..intg..times..intg..times..function..times.d.times.d.times..intg..-
times..intg..times..times..function..alpha..times..times..times..times.d.t-
imes.d.times..times..times..intg..times..intg..times..alpha..times..times.-
.times.d.times.d.times..times..times..times..times..times..alpha..times..t-
imes..alpha..function..alpha..function..times..alpha..times..times..alpha.-
.times..times. ##EQU00059##
That is to say, according to this processing, the error calculation
unit 823 serves as, so to speak, a kind of pixel value generating
unit, and generates pixel values from the approximation
function.
In step S825, the error calculation unit 803 calculates the
difference between a pixel value obtained with integration such as
shown in the above Expression (92) and a pixel value of the input
image, and outputs this to the comparison unit 824 as an error.
That is to say, the error calculation unit 823 obtains the
difference between the pixel value of a pixel corresponding to the
integral range shown in the above FIG. 171 and FIG. 172 (T.sub.m
through T.sub.m+1 for the spatial direction T, and y.sub.m through
y.sub.m+1 for the spatial direction Y) and a pixel value obtained
by the integrated result in a range corresponding to the pixel, as
an error, and outputs this to the comparison unit 824.
In step S826, the comparison unit 824 determines regarding whether
or not the absolute value of the error between the pixel value
obtained with integration and the pixel value of the input image,
which are input from the error calculation unit 823, is a
predetermined threshold value or less.
In step S826, in the event that determination is made that the
error is the predetermined threshold value or less, since the pixel
value obtained with integration is a value close to the pixel value
of the input image, the comparison unit 824 regards the
approximation function set for calculating the pixel value of the
pixel as a function sufficiently approximated with the light
intensity allocation of a light signal in the real world, and
recognizes the region of the pixel now processed as a processing
region in step S827. In further detail, the comparison unit 824
stores the pixel now processed in unshown memory as the pixel in
the subsequent processing regions.
On the other hand, in the event that determination is made that the
error is not the threshold value or less in step S826, since the
pixel value obtained with integration is a value far from the
actual pixel value, the comparison unit 824 regards the
approximation function set for calculating the pixel value of the
pixel as a function insufficiently approximated with the light
intensity allocation in the real world, and recognizes the region
of the pixel now processed as a non-processing region where
processing using the approximation function based on continuity
information is not performed at a subsequent stage in step S828. In
further detail, the comparison unit 824 stores the region of the
pixel now processed in unshown memory as the subsequent
non-processing regions.
In step S829, the comparison unit 824 determines regarding whether
or not the processing has been performed as to all of the pixels,
and in the event that determination is made that the processing has
not been performed as to all of the pixels, the processing returns
to step S822, wherein the subsequent processing is repeatedly
performed. In other words, the processing in steps S822 through
S829 is repeatedly performed until determination processing wherein
comparison between a pixel value obtained with integration and a
pixel value input is performed, and determination is made regarding
whether or not the pixel is a processing region, is completed
regarding all of the pixels.
In step S829, in the event that determination is made that
determination processing wherein comparison between a pixel value
obtained by reintegration and a pixel value input is performed, and
determination is made regarding whether or not the pixel is a
processing region, has been completed regarding all of the pixels,
the comparison unit 824, in step S830, outputs region information
wherein a processing region where processing based on the
continuity information in the frame direction is performed at
subsequent processing, and a non-processing region where processing
based on the continuity information in the frame direction is not
performed are identified regarding the input image stored in the
unshown memory, as continuity information.
According to the above processing, based on the error between the
pixel value obtained by the integrated result in a region
corresponding to each pixel using the approximation function f(t)
calculated based on the continuity information and the pixel value
within the actual input image, evaluation for reliability of
expression of the approximation function is performed for each
region (for each pixel), and accordingly, a region having a small
error, i.e., only a region where a pixel of which the pixel value
obtained with integration based on the approximation function is
reliable exists is regarded as a processing region, and the regions
other than this region are regarded as non-processing regions, and
consequently, only a reliable region can be subjected to the
processing based on continuity information in the frame direction,
and the necessary processing alone can be performed, whereby
processing speed can be improved, and also the processing can be
performed as to the reliable region alone, resulting in preventing
image quality due to this processing from deterioration.
An arrangement may be made wherein the configurations of the data
continuity information detecting unit 101 in FIG. 165 and FIG. 169
are combined, any one-dimensional direction of the spatial and
temporal directions is selected, and the region information is
selectively output.
According to the above configuration, light signals in the real
world are projected by the multiple detecting elements of the
sensor each having spatio-temporal integration effects, continuity
of data in image data made up of multiple pixels having a pixel
value projected by the detecting elements of which a part of
continuity of the light signals in the real world drops is
detected, a function corresponding to the light signals in the real
world is approximated on condition that the pixel value of each
pixel corresponding to the detected continuity, and corresponding
to at least a position in a one-dimensional direction of the
spatial and temporal directions of the image data is the pixel
value acquired with at least integration effects in the
one-dimensional direction, and accordingly, a difference value
between a pixel value acquired by estimating the function
corresponding to the light signals in the real world, and
integrating the estimated function at least in increments of
corresponding to each pixel in the primary direction and the pixel
value of each pixel is detected, and the function is selectively
output according to the difference value, and accordingly, a region
alone where a pixel of which the pixel value obtained with
integration based on the approximation function is reliable exists
can be regarded as a processing region, and the other regions other
than this region can be regarded as non-processing regions, the
reliable region alone can be subjected to processing based on the
continuity information in the frame direction, so the necessary
processing alone can be performed, whereby processing speed can be
improved, and also the reliable region alone can be subjected to
processing, resulting in preventing image quality due to this
processing from deterioration.
Next, description will be made regarding a continuity detecting
unit 101 wherein angle as continuity can be obtained more
accurately and also at higher speed with reference to FIG. 173.
An simple-type angle detecting unit 901 is essentially the same as
the continuity detecting unit 101 described with reference to FIG.
95, compares a block corresponding to the pixel of interest with a
perimeter pixel block around the pixel of interest to detect an
angle range between the pixel of interest and the perimeter pixel
wherein the correlation between the block corresponding to the
pixel of interest and the perimeter pixel block is strongest, which
is so-called block matching, thereby simply detecting regarding
which range of 16 directions (e.g., in the case in which an angle
of data continuity is taken as .theta., 16 ranges of
0.ltoreq..theta.<18.4, 18.4.ltoreq..theta.<26.05,
26.05.ltoreq..theta.<33.7, 33.7.ltoreq..theta.<45,
45.ltoreq..theta.<56.3, 56.3.ltoreq..theta.<63.95,
63.95.ltoreq..theta.<71.6, 71.6.ltoreq..theta.<90,
90.ltoreq..theta.<108.4, 108.4.ltoreq..theta.<116.05,
116.05.ltoreq..theta.<123.7, 123.7.ltoreq..theta.<135,
135.ltoreq..theta.<146.3, 146.3.ltoreq..theta.<153.95,
153.95.ltoreq..theta.<161.6, and 161.6.ltoreq..theta.<180 in
FIG. 178 described later) an angle as continuity belongs to, and
outputting each median (or a representing value within the range
thereof) to a determining unit 902.
The determining unit 902 determines, based on the angle as
continuity information simply obtained, which is input from the
simple-type angle detecting unit 901, regarding whether the input
angle is an angle closer to the vertical direction, or an angle
closer to the horizontal direction, or other than those, and
controls a switch 903 to connect to any one of terminals 903a and
903b according to the determination result to supply an input image
to a regression-type angle detecting unit 904 or a gradient-type
angle detecting unit 905, and also supplies the angle information
simply obtained, which is input from the simple-type angle
detecting unit 901, to the regression-type angle detecting unit 904
when the switch 903 is connected to the terminal 903a.
More particularly, in the event that the determining unit 902
determines that the direction of continuity supplied from the
simple-type angle detecting unit 901 is an angle closer to the
horizontal direction or the vertical direction (e.g., in the event
that the angle .theta. of the continuity input from the simple-type
angle detecting unit 901 is 0.ltoreq..theta..ltoreq.18.4,
71.6.ltoreq..theta..ltoreq.108.4, or
161.6.ltoreq..theta..ltoreq.180), the determining unit 902 controls
the switch 903 to connect to the terminal 903a to supply the input
image to the regression-type angle detecting unit 904, and in the
event of other than that, i.e., in the event that the direction of
continuity is closer to 45 degrees or 135 degrees, the determining
unit 902 controls the switch 903 to connect to the terminal 903b to
supply the input image to the gradient-type angle detecting unit
905.
The regression-type angle detecting unit 904 has a configuration
essentially similar to the continuity detecting unit 101 described
with reference to the above FIG. 107, regressively (in the event
that the correlation value between the pixel value of the pixel of
interest and the pixel value of a pixel belonged to a region
corresponding to the pixel of interest is equal to or greater than
a threshold value, the score according to the correlation value is
set to such a pixel, whereby the score of the pixel belonged to the
region is detected, and also the angle of data continuity is
obtained by a regression line detected based on the detected score)
performs detection of an angle of data continuity, and outputs the
detected angle to the actual world estimating unit 102 as data
continuity information. However, when the regression-type angle
detecting unit 904 detects the angle, the regression-type angle
detecting unit 904 restricts a range (scope) corresponding to the
pixel of interest, sets a score, and regressively detects an angle,
based on the angle supplied from the determining unit 902.
The gradient-type angle detecting unit 905 is essentially similar
to the continuity detecting unit 101 described with reference to
FIG. 124, detects an angle of data continuity based on the
difference between the maximum value and minimum value of the pixel
values of the block corresponding to the pixel of interest (the
above dynamic range block), i.e., the dynamic range (essentially,
based on the gradient between the maximum value and minimum value
of the pixels in the dynamic range block), and outputs this angle
to the actual world estimating unit 102 as data continuity
information.
Next, description will be made regarding the configuration of the
simple-type angle detecting unit 901 with reference to FIG. 174,
but the simple-type angle detecting unit 901 has essentially the
same configuration as that of the data continuity detecting unit
101 described with reference to FIG. 95. Accordingly, a data
selecting unit 911, error estimating unit 912, continuity direction
derivation unit 913, pixel selecting units 921-1 through 921-L,
estimating error calculating units 922-1 through 922-L, and
smallest error angle selecting unit 923 of the simple-type angle
detecting unit 901 shown in FIG. 174 are similar to the data
selecting unit 441, error estimating unit 442, continuity direction
derivation unit 443, pixel selecting units 461-1 through 461-L,
estimating error calculating units 462-1 through 462-L, and
smallest error angle selecting unit 443 of the data continuity
detecting unit 101 shown in FIG. 95, so description of each unit is
omitted.
Next, description will be made regarding the configuration of the
regression-type angle detecting unit 904 with reference to FIG.
175, but the regression-type angle detecting unit 904 has
essentially the same configuration as that of the data continuity
detecting unit 101 described with reference to FIG. 107.
Accordingly, frame memory 931, pixel acquiring unit 932, regression
line computing unit 934, and angle calculating unit 935 of the
regression-type angle detecting unit 904 shown in FIG. 175 are
similar to the frame memory 501, pixel acquiring unit 502,
regression line computing unit 504, and angle calculating unit 505
of the data continuity detecting unit 101 shown in FIG. 107, so
description thereof will be omitted.
Here, the difference in the regression-type angle detecting unit
904 as to the data continuity detecting unit 101 shown in FIG. 107
is a score detecting unit 933. The score detecting unit 933 has the
same function as the score detecting unit 503 shown in FIG. 107,
but further includes scope memory 933a, detects a score based on
angle range information which detects a score corresponding to the
pixel of interest stored in the scope memory 933a based on the
angle of data continuity detected by the simple-type angle
detecting unit 901 input from the determining unit 902, and
supplies the detected score information to the regression line
computing unit 934.
Next, description will be made regarding the configuration of the
gradient-type angle detecting unit 905 with reference to FIG. 176,
but the gradient-type angle detecting unit 905 has essentially the
same configuration as that of the data continuity detecting unit
101 described with reference to FIG. 124. Accordingly, a data
selecting unit 941, data supplementing unit 942, continuity
direction derivation unit 943, horizontal/vertical determining unit
951, data acquiring unit 952, difference supplementing unit 961,
MaxMin acquiring unit 962, difference supplementing unit 963, and
continuity direction computation unit 971 shown in FIG. 176 are
similar to the data selecting unit 701, data supplementing unit
702, continuity direction derivation unit 703, horizontal/vertical
determining unit 711, data acquiring unit 712, difference
supplementing unit 721, MaxMin acquiring unit 722, difference
supplementing unit 723, and continuity direction computation unit
731 of the data continuity detecting unit 101 shown in FIG. 124, so
description thereof will be omitted.
Next, description will be made regarding the processing for
detecting data continuity with reference to the flowchart in FIG.
177.
In step S901, the simple-type angle detecting unit 901 executes the
simple-type angle detecting processing, and outputs the detected
angle information to the determining unit 902. Note that the
simple-type angle detecting processing is the same as the
processing for detecting data continuity described with reference
to the flowchart in FIG. 103, so description thereof will be
omitted.
In step S902, the determining unit 902 determines regarding whether
an angle of data continuity is closer to the horizontal direction
or the vertical direction based on the angle information of data
continuity input from the simple-type angle detecting unit 901.
More particularly, the determining unit 902 determines that the
angle of data continuity is closer to the horizontal direction or
the vertical direction in the event that the angle of data
continuity, i.e., the angle .theta. of continuity input from the
simple-type angle detecting unit 901 is in a range of
0.ltoreq..theta..ltoreq.18.4, 71.6.ltoreq..theta..ltoreq.108.4, or
161.6.ltoreq..theta..ltoreq.180, for example.
In step S902, in the event that determination is made that the
angle of data continuity is the horizontal direction or the
vertical direction, the processing proceeds to step S903.
In step S903, the determining unit 902 controls the switch 903 to
connect to the terminal 903a, and also supplies the angle
information of data continuity supplied from the simple-type angle
detecting unit 901 to the regression-type angle detecting unit 904.
According to this processing, the input image and the angle
information of data continuity detected by the simple-type angle
detecting unit 901 are supplied to the regression-type angle
detecting unit 904.
In step S904, the regression-type angle detecting unit 904 executes
the regression-type angle detecting processing, and outputs the
detected angle to the actual world estimating unit 102 as data
continuity information. Note that description will be made later
regarding the regression-type angle detecting processing with
reference to FIG. 179.
In step S905, the data selecting unit 911 of the simple-type angle
detecting unit 901 determines regarding whether or not the
processing has been completed regarding all of the pixels, and in
the event that determination is made that the processing has not
been completed regarding all of the pixels, the processing returns
to step S901, wherein the subsequent processing is repeatedly
performed.
On the other hand, in the event that determination is made that the
direction of data continuity is not the horizontal direction nor
the vertical direction in step S902, the processing proceeds to
step S906.
In step S906, the determining unit 902 controls the switch 903 to
connect to the terminal 903b. According to this processing, an
input image is supplied to the gradient-type angle detecting unit
905.
In step S907, the gradient-type angle detecting unit 905 executes
the gradient-type angle detecting processing to detect an angle,
and outputs the detected angle to the actual world estimating unit
102 as continuity information. Note that the gradient-type angle
detecting processing is essentially the same processing as the
processing for detecting data continuity described with reference
to the flowchart in FIG. 149, so description thereof will be
omitted.
That is to say, when determination is made in the processing of
step S902 that the angle of data continuity detected by the
simple-type angle detecting unit 901 is the angle corresponding to
a white region with no slant line (18.4.ltoreq..theta..ltoreq.71.6,
or 108.4.ltoreq..theta..ltoreq.161.6) in the event that the pixel
of interest is the center in the drawing as shown in FIG. 178, the
determining unit 902 controls the switch 903 to connect to the
terminal 903a in the processing of step S903, whereby the
regression-type angle detecting unit 904 obtains a regression
straight line using correlation to detect an angle of data
continuity from the regression line in the processing of step
S904.
Also, when determination is made in the processing of step S902
that the angle of data continuity detected by the simple-type angle
detecting unit 901 is the angle corresponding to the region of a
slant line portion (0.ltoreq..theta..ltoreq.18.4,
71.6.ltoreq..theta..ltoreq.108.4, or
161.6.ltoreq..theta..ltoreq.180) in the event that the pixel of
interest is the center in the drawing as shown in FIG. 178, the
determining unit 902 controls the switch 903 to connect to the
terminal 903b in the processing of step S906, whereby the
gradient-type angle detecting unit 905 detects an angle of data
continuity in the processing of step S907.
The regression-type angle detecting unit 904 compares the
correlation between a block corresponding to the pixel of interest
and a block corresponding to a perimeter pixel, and obtains an
angle of data continuity from the angle as to the pixel
corresponding to the block having the strongest correlation.
Accordingly, in the event that the angle of data continuity is
closer to the horizontal direction or the vertical direction, there
is a possibility that the pixel belonged to the block having the
strongest correlation is far away from the pixel of interest, so
that a search region needs to be expanded in order to detect the
block of a strong-correlation perimeter pixel accurately, resulting
in the threat of vast processing, and further expanding a search
region allows the threat of accidentally detecting a block
strong-correlated with the block corresponding to the pixel of
interest in a position where continuity does not exist actually,
and allows the threat of deteriorating the detection accuracy of an
angle.
Conversely, with the gradient-type angle detecting unit 905, the
closer to the horizontal direction or the vertical direction the
angle of data continuity is, the farther the distance between the
pixels that take the maximum value and minimum value of pixel
values within a dynamic range block is apart from each other,
resulting in increase of pixels having the same gradient (gradient
indicating change in pixel values) within the extracted block, and
accordingly, performing statistical processing enables the angle of
data continuity to be detected more accurately.
On the other hand, with the gradient-type angle detecting unit 905,
the closer to 45 degrees or 135 degrees the angle of data
continuity is, the closer the distance between the pixels that take
the maximum value and minimum value of pixel values within a
dynamic range block is, resulting in decrease of pixels having the
same gradient (gradient indicating change in pixel values) within
the extracted block, and accordingly, performing statistical
processing deteriorates the accuracy of the angle of data
continuity.
Conversely, with the regression-type angle detecting unit 904, in
the event that the angle of data continuity is around 45 degrees or
135 degrees, a block corresponding to the pixel of interest and a
block corresponding to a strong-correlation pixel exist with a
short distance, whereby the angle of continuity can be detected
more accurately.
Consequently, an angle of data continuity can be detected more
accurately in all of ranges by switching the processing based on
the angle detected by the simple-type angle detecting unit 901,
according to each property of the regression-type angle detecting
unit 904 and gradient-type angle detecting unit 905. Further, an
angle of data continuity can be detected accurately, so that the
actual world can be estimated accurately, and consequently, a more
accurate and higher-precision (image) processing result can be
obtained as to events in the real world.
Next, the regression-type angle detecting processing, which is the
processing of step S904 in the flowchart in FIG. 177, will be
described with reference to the flowchart shown in FIG. 179.
Note that the regression-type angle detecting processing using the
regression-type angle detecting unit 904 is similar to the
processing for detecting data continuity described with reference
to the flowchart shown in FIG. 114, the processing of steps S921
through S922 and S924 through S927 in the flowchart shown in FIG.
179 are the same as the processing of steps through S501 through
S506 in the flowchart shown in FIG. 114, so description thereof
will be omitted.
In step S923, the score detecting unit 933 rejects pixels other
than a scope range from the pixels to be processed with reference
to the scope memory 933a, based on the angle information of data
continuity detected by the simple-type angle detecting unit 901
supplied from the determining unit 902.
That is to say, for example, in the event that the range of the
angle .theta. detected by the simple-type angle detecting unit 901
is 45.ltoreq..theta..ltoreq.56.3, a pixel range corresponding to
the slant portion shown in FIG. 180 is stored in the scope memory
933a as scope corresponding to the range, and the score detecting
unit 933 rejects pixels other than the range corresponding to the
scope from the range to be processed.
As for a more detailed example of a scope range corresponding to
each angle, for example, in the event that the angle of data
continuity detected by the simple-type angle detecting unit 901 is
50 degrees, the image in the scope range and the pixels other than
the scope range are defined beforehand, as shown in FIG. 181. Note
that FIG. 181 illustrates an example in the case of a range of 31
pixels.times.31 pixels centered on the pixel of interest, each
allocation shown with 0 and 1 indicates a pixel position, and a
position surrounded with a circle mark in the center of the drawing
is the position of the pixel of interest. Also, the pixels in the
position shown with 1 are pixels within a scope range, and the
pixels in the position shown with 0 are pixels other than the scope
range. Note that the above description is applicable to the
following FIG. 182 through FIG. 183.
That is to say, the pixels serving as the scope range are disposed
centered on the pixel of interest along around angle 50 degrees
with a certain range width, as shown in FIG. 181.
Also, in the same way, in the event that the angle detected by the
simple-type angle detecting unit 901 is 60 degrees, the pixels
serving as the scope range are disposed centered on the pixel of
interest along around angle 60 degrees with a certain range width,
as shown in FIG. 182.
Further, in the event that the angle detected by the simple-type
angle detecting unit 901 is 67 degrees, the pixels serving as the
scope range are disposed centered on the pixel of interest along
around angle 67 degrees with a certain range width, as shown in
FIG. 183.
Also, in the event that the angle detected by the simple-type angle
detecting unit 901 is 81 degrees, the pixels serving as the scope
range are disposed centered on the pixel of interest along around
angle 81 degrees with a certain range width, as shown in FIG.
184.
As described above, the pixels other than the scope range are
rejected from the range to be processed, so that the processing of
pixels existing in positions away from continuity of data can be
omitted at the processing for converting each pixel value, which is
the processing of step S924, into a score, and consequently, the
strong-correlated pixels alone along the direction of data
continuity, which are to be processed, are processed, thereby
improving the processing speed. Further, scores can be obtained
using the strong-correlated pixels alone along the direction of
data continuity, which are to be processed, whereby an angle of
data continuity can be detected more accurately.
Note that the pixels belonged to the scope range are not restricted
to the ranges shown in FIG. 181 through FIG. 184, rather, a range
having a various width present in a position along the angle
detected by the simple-type angle detecting unit 901, which is made
up of multiple pixels centered on the pixel of interest, may be
employed.
Also, with the data continuity detecting unit 101 described with
reference to FIG. 173, the determining unit 902 controls the switch
903 based on the angle information of data continuity detected by
the simple-type angle detecting unit 901 to input the input image
information to either the regression-type angle detecting unit 904
or the gradient-type angle detecting unit 905, but an arrangement
may be made wherein the input image is input to both the
regression-type angle detecting unit 904 and the gradient-type
angle detecting unit 905, the angle detecting processing is
performed on the both units, following which the angle information
detected in any one of the processing is output based on the angle
information of data continuity detected by the simple-type angle
detecting unit 901.
FIG. 185 illustrates the configuration of the data continuity
detecting unit 101, which is configured such that the input image
is input to both the regression-type angle detecting unit 904 and
the gradient-type angle detecting unit 905, the angle detecting
processing is performed on the both units, following which the
angle information detected in any one of the processing is output
based on the angle information of data continuity detected by the
simple-type angle detecting unit 901. Note that the same components
as the data continuity detecting unit 101 shown in FIG. 173 are
denoted with the same reference numerals, so description thereof
will be omitted as appropriate.
With the configuration of the data continuity detecting unit 101
shown in FIG. 185, the difference as to the data continuity
detecting unit 101 shown in FIG. 173 is in that the switch 903 is
deleted, input image data is input to both the regression-type
angle detecting unit 904 and the gradient-type angle detecting unit
905, on each output side thereof a switch 982 is provided
respectively, the angle information detected with either method is
output by switching connection of each terminal 982a or 982b
thereof respectively. Note that the switch 982 shown in FIG. 185 is
essentially the same as the switch 903 shown in FIG. 173, so
description thereof will be omitted.
Next, description will be made regarding data continuity detection
processing using the data continuity detecting unit 101 in FIG. 185
with reference to the flowchart in FIG. 186. Note that the
processing of steps S941, S943 through S945, and S947 in the
flowchart shown in FIG. 186 is the same as the processing of steps
S901, S904, S907, S902, and S905 shown in FIG. 177, so description
thereof will be omitted.
In step S942, the determining unit 902 outputs the angle
information of data continuity input from the simple-type angle
detecting unit 901 to the regression-type angle detecting unit
904.
In step S946, the determining unit 902 controls the switch 982 to
connect to the terminal 982a.
In step S948, the determining unit 902 controls the switch 982 to
connect to the terminal 982b.
Note that with the flowchart shown in FIG. 186, the processing
order of steps S943 and S944 may be exchanged.
According to the above arrangement, the simple-type angle detecting
unit 901 detects an angle corresponding to the reference axis of
continuity of image data in image data made up of a plurality of
pixels acquired by real world light signals being cast upon a
plurality of detecting elements each having spatio-temporal
integration effects, of which a part of continuity of the real
world light signals have been lost, using the matching processing,
and the regression-type angle detecting unit 904 or the
gradient-type angle detecting unit 905 detects, based on the image
data within a predetermined region corresponding to the detected
angle, an angle using the statistical processing, thereby detecting
an angle of data continuity more accurately at higher speed.
Next, description will be made regarding estimation of signals in
the actual world 1.
FIG. 187 is a block diagram illustrating the configuration of the
actual world estimating unit 102.
With the actual world estimating unit 102 of which the
configuration is shown in FIG. 187, based on the input image and
the data continuity information supplied from the continuity
detecting unit 101, the width of a fine line in the image, which is
a signal in the actual world 1, is detected, and the level of the
fine line (light intensity of the signal in the actual world 1) is
estimated.
A line-width detecting unit 2101 detects the width of a fine line
based on the data continuity information indicating a continuity
region serving as a fine-line region made up of pixels, on which
the fine-line image is projected, supplied from the continuity
detecting unit 101. The line-width detecting unit 2101 supplies
fine-line width information indicating the width of a fine line
detected to a signal-level estimating unit 2102 along with the data
continuity information.
The signal-level estimating unit 2102 estimates, based on the input
image, the fine-line width information indicating the width of a
fine line, which is supplied from the line-width detecting unit
2101, and the data continuity information, the level of the
fine-line image serving as the signals in the actual world 1, i.e.,
the level of light intensity, and outputs actual world estimating
information indicating the width of a fine line and the level of
the fine-line image.
FIG. 188 and FIG. 189 are diagrams for describing processing for
detecting the width of a fine line in signals in the actual world
1.
In FIG. 188 and FIG. 189, a region surrounded with a thick line
(region made up of four squares) denotes one pixel, a region
surrounded with a dashed line denotes a fine-line region made up of
pixels on which a fine-line image is projected, and a circle
denotes the gravity of a fine-line region. In FIG. 188 and FIG.
189, a hatched line denotes a fine-line image cast in the sensor 2.
In other words, it can be said that this hatched line denotes a
region where a fine-line image in the actual world 1 is projected
on the sensor 2.
In FIG. 188 and FIG. 189, S denotes a gradient to be calculated
from the gravity position of a fine-line region, and D is the
duplication of fine-line regions. Here, fine-line regions are
adjacent to each other, so the gradient S is a distance between the
gravities thereof in increments of pixel. Also, the duplication D
of fine-line regions denotes the number of pixels adjacent to each
other in two fine-line regions.
In FIG. 188 and FIG. 189, W denotes the width of a fine line.
In FIG. 188, the gradient S is 2, and the duplication D is 2.
In FIG. 189, the gradient S is 3, and the duplication D is 1.
The fine-line regions are adjacent to each other, and the distance
between the gravities thereof in the direction where the fine-line
regions are adjacent to each other is one pixel, so W:D=1:S holds,
the fine-line width W can be obtained by the duplication D/gradient
S.
For example, as shown in FIG. 188, when the gradient S is 2, and
the duplication D is 2, 2/2 is 1, so the fine-line width W is 1.
Also, for example, as shown in FIG. 189, when the gradient S is 3,
and the duplication D is 1, the fine-line width W is 1/3.
The line-width detecting unit 2101 thus detects the width of a
fine-line based on the gradient calculated from the gravity
positions of fine-line regions, and duplication of fine-line
regions.
FIG. 190 is a diagram for describing the processing for estimating
the level of a fine-line signal in signals in the actual world
1.
In FIG. 190, a region surrounded with a thick line (region made up
of four squares) denotes one pixel, a region surrounded with a
dashed line denotes a fine-line region made up of pixels on which a
fine-line image is projected. In FIG. 190, E denotes the length of
a fine-line region in increments of a pixel in a fine-line region,
and D is duplication of fine-line regions (the number of pixels
adjacent to another fine-line region).
The level of a fine-line signal is approximated when the level is
constant within processing increments (fine-line region), and the
level of an image other than a fine line wherein a fine line is
projected on the pixel value of a pixel is approximated when the
level is equal to a level corresponding to the pixel value of the
adjacent pixel.
With the level of a fine-line signal as C, let us say that with a
signal (image) projected on the fine-line region, the level of the
left side portion of a portion where the fine-line signal is
projected is A in the drawing, and the level of the right side
portion of the portion where the fine-line signal is projected is B
in the drawing.
At this time, Expression (93) holds. Sum of pixel values of a
fine-line region=(E-D)/2.times.A+(E-D)/2.times.B+D.times.C (93)
The width of a fine line is constant, and the width of a fine-line
region is one pixel, so the area of (the portion where the signal
is projected of) a fine line in a fine-line region is equal to the
duplication D of fine-line regions. The width of a fine-line region
is one pixel, so the area of a fine-line region in increments of a
pixel in a fine-line region is equal to the length E of a fine-line
region.
Of a fine-line region, the area on the left side of a fine line is
(E-D)/2. Of a fine-line region, the area on the right side of a
fine line is (E-D)/2.
The first term of the right side of Expression (93) is the portion
of the pixel value where the signal having the same level as that
in the signal projected on a pixel adjacent to the left side is
projected, and can be represented with Expression (94).
.alpha..times..times..times. ##EQU00060##
In Expression (94), A.sub.i denotes the pixel value of a pixel
adjacent to the left side.
In Expression (94), .alpha.i denotes the proportion of the area
where the signal having the same level as that in the signal
projected on a pixel adjacent to the left side is projected on the
pixel of the fine-line region. In other words, .alpha..sub.i
denotes the proportion of the same pixel value as that of a pixel
adjacent to the left side, which is included in the pixel value of
the pixel in the fine-line region.
i represents the position of a pixel adjacent to the left side of
the fine-line region.
For example, in FIG. 190, the proportion of the same pixel value as
the pixel value A.sub.0 of a pixel adjacent to the left side of the
fine-line region, which is included in the pixel value of the pixel
in the fine-line region, is .alpha..sub.0. In FIG. 190, the
proportion of the same pixel value as the pixel value A.sub.1 of a
pixel adjacent to the left side of the fine-line region, which is
included in the pixel value of the pixel in the fine-line region,
is .alpha..sub.1. In FIG. 190, the proportion of the same pixel
value as the pixel value A.sub.2 of a pixel adjacent to the left
side of the fine-line region, which is included in the pixel value
of the pixel in the fine-line region, is .alpha..sub.2.
The second term of the right side of Expression (93) is the portion
of the pixel value where the signal having the same level as that
in the signal projected on a pixel adjacent to the right side is
projected, and can be represented with Expression (95).
.beta..times..times..times. ##EQU00061##
In Expression (95), B.sub.j denotes the pixel value of a pixel
adjacent to the right side.
In Expression (95), .beta..sub.j denotes the proportion of the area
where the signal having the same level as that in the signal
projected on a pixel adjacent to the right side is projected on the
pixel of the fine-line region. In other words, .beta..sub.j denotes
the proportion of the same pixel value as that of a pixel adjacent
to the right side, which is included in the pixel value of the
pixel in the fine-line region.
j denotes the position of a pixel adjacent to the right side of the
fine-line region.
For example, in FIG. 190, the proportion of the same pixel value as
the pixel value B.sub.0 of a pixel adjacent to the right side of
the fine-line region, which is included in the pixel value of the
pixel in the fine-line region, is .beta..sub.0. In FIG. 190, the
proportion of the same pixel value as the pixel value B.sub.1 of a
pixel adjacent to the right side of the fine-line region, which is
included in the pixel value of the pixel in the fine-line region,
is .beta..sub.1. In FIG. 190, the proportion of the same pixel
value as the pixel value B.sub.2 of a pixel adjacent to the right
side of the fine-line region, which is included in the pixel value
of the pixel in the fine-line region, is .beta..sub.2.
Thus, the signal level estimating unit 2102 obtains the pixel
values of the image including a fine line alone, of the pixel
values included in a fine-line region, by calculating the pixel
values of the image other than a fine line, of the pixel values
included in the fine-line region, based on Expression (94) and
Expression (95), and removing the pixel values of the image other
than the fine line from the pixel values in the fine-line region
based on Expression (93). Subsequently, the signal level estimating
unit 2102 obtains the level of the fine-line signal based on the
pixel values of the image including the fine line alone and the
area of the fine line. More specifically, the signal level
estimating unit 2102 calculates the level of the fine line signal
by dividing the pixel values of the image including the fine line
alone, of the pixel values included in the fine-line region, by the
area of the fine line in the fine-line region, i.e., the
duplication D of the fine-line regions.
The signal level estimating unit 2102 outputs actual world
estimating information indicating the width of a fine line, and the
signal level of a fine line, in a signal in the actual world 1.
With the technique of the present invention, the waveform of a fine
line is geometrically described instead of pixels, so any
resolution can be employed.
Next, description will be made regarding actual world estimating
processing corresponding to the processing in step S102 with
reference to the flowchart in FIG. 191.
In step S2101, the line-width detecting unit 2101 detects the width
of a fine line based on the data continuity information. For
example, the line-width detecting unit 2101 estimates the width of
a fine line in a signal in the actual world 1 by dividing
duplication of fine-line regions by a gradient calculated from the
gravity positions in fine-line regions.
In step S2102, the signal level estimating unit 2102 estimates the
signal level of a fine line based on the width of a fine line, and
the pixel value of a pixel adjacent to a fine-line region, outputs
actual world estimating information indicating the width of the
fine line and the signal level of the fine line, which are
estimated, and the processing ends. For example, the signal level
estimating unit 2102 obtains pixel values on which the image
including a fine line alone is projected by calculating pixel
values on which the image other than the fine line included in a
fine-line region is projected, and removing the pixel values on
which the image other than the fine line from the fine-line region
is projected, and estimates the level of the fine line in a signal
in the actual world 1 by calculating the signal level of the fine
line based on the obtained pixel values on which the image
including the fine line alone is projected, and the area of the
fine line.
Thus, the actual world estimating unit 102 can estimate the width
and level of a fine line of a signal in the actual world 1.
As described above, a light signal in the real world is projected,
continuity of data regarding first image data wherein part of
continuity of a light signal in the real world drops, is detected,
the waveform of the light signal in the real world is estimated
from the continuity of the first image data based on a model
representing the waveform of the light signal in the real world
corresponding to the continuity of data, and in the event that the
estimated light signal is converted into second image data, a more
accurate higher-precision processing result can be obtained as to
the light signal in the real world.
FIG. 192 is a block diagram illustrating another configuration of
the actual world estimating unit 102.
With the actual world estimating unit 102 of which the
configuration is illustrated in FIG. 192, a region is detected
again based on an input image and the data continuity information
supplied from the data continuity detecting unit 101, the width of
a fine line in the image serving as a signal in the actual world 1
is detected based on the region detected again, and the light
intensity (level) of the signal in the actual world 1 is estimated.
For example, with the actual world estimating unit 102 of which the
configuration is illustrated in FIG. 192, a continuity region made
up of pixels on which a fine-line image is projected is detected
again, the width of a fine line in an image serving as a signal in
the actual world 1 is detected based on the region detected again,
and the light intensity of the signal in the actual world 1 is
estimated.
The data continuity information, which is supplied from the data
continuity detecting unit 101, input to the actual world estimating
unit 102 of which configuration is shown in FIG. 192, includes
non-continuity component information indicating non-components
other than continuity components on which a fine-line image is
projected, of input images serving as the data 3, monotonous
increase/decrease region information indicating a monotonous
increase/decrease region of continuity regions, information
indicating a continuity region, and the like. For example,
non-continuity component information included in the data
continuity information is made up of the gradient of a plane and
intercept which approximate non-continuity components such as a
background in an input image.
The data continuity information input to the actual world
estimating unit 102 is supplied to a boundary detecting unit 2121.
The input image input to the actual world estimating unit 102 is
supplied to the boundary detecting unit 2121 and signal level
estimating unit 2102.
The boundary detecting unit 2121 generates an image made up of
continuity components alone on which a fine-line image is projected
from the non-continuity component information included in the data
continuity information, and the input image, calculates an
allocation ratio indicating a proportion wherein a fine-line image
serving as a signal in the actual world 1 is projected, and detects
a fine-line region serving as a continuity region again by
calculating a regression line indicating the boundary of the
fine-line region from the calculated allocation ratio.
FIG. 193 is a block diagram illustrating the configuration of the
boundary detecting unit 2121.
An allocation-ratio calculation unit 2131 generates an image made
up of continuity components alone on which a fine-line image is
projected from the data continuity information, the non-continuity
component information included in the data continuity information,
and an input image. More specifically, the allocation-ratio
calculation unit 2131 detects adjacent monotonous increase/decrease
regions of the continuity region from the input image based on the
monotonous increase/decrease region information included in the
data continuity information, and generates an image made up of
continuity components alone on which a fine-line image is projected
by subtracting an approximate value to be approximated at a plane
indicated with a gradient and intercept included in the continuity
component information from the pixel value of a pixel belonged to
the detected monotonous increase/decrease region.
Note that the allocation-ratio calculation unit 2131 may generate
an image made up of continuity components alone on which a
fine-line image is projected by subtracting an approximate value to
be approximated at a plane indicated with a gradient and intercept
included in the continuity component information from the pixel
value of a pixel in the input image.
The allocation-ratio calculation unit 2131 calculates an allocation
ratio indicating proportion wherein a fine-line image serving as a
signal in the actual world 1 is allocated into two pixels belonged
to adjacent monotonous increase/decrease regions within a
continuity region based on the generated image made up of the
continuity components alone. The allocation-ratio calculation unit
2131 supplies the calculated allocation ratio to a regression-line
calculation unit 2132.
Description will be made regarding allocation-ratio calculation
processing in the allocation-ratio calculation unit 2131 with
reference to FIG. 194 through FIG. 196.
The numeric values in two columns on the left side in FIG. 194
denote the pixel values of pixels vertically arrayed in two columns
of an image calculated by subtracting approximate values to be
approximated at a plane indicated with a gradient and intercept
included in the continuity component information from the pixel
values of an input image. Two regions surrounded with a square on
the left side in FIG. 194 denote a monotonous increase/decrease
region 2141-1 and monotonous increase/decrease region 2141-2, which
are two adjacent monotonous increase/decrease regions. In other
words, the numeric values shown in the monotonous increase/decrease
region 2141-1 and monotonous increase/decrease region 2141-2 denote
the pixel values of pixels belonged to a monotonous
increase/decrease region serving as a continuity region, which is
detected by the data continuity detecting unit 101.
The numeric values in one column on the right side in FIG. 194
denote values obtained by adding the pixel values of the pixels
horizontally arrayed, of the pixel values of the pixels in two
columns on the left side in FIG. 194. In other words, the numeric
values in one column on the right side in FIG. 194 denote values
obtained by adding the pixel values on which a fine-line image is
projected for each pixel horizontally adjacent regarding the two
monotonous increase/decrease regions made up of pixels in one
column vertically arrayed.
For example, when belonging to any one of the monotonous
increase/decrease region 2141-1 and monotonous increase/decrease
region 2141-2, which are made up of the pixels in one column
vertically arrayed respectively, and the pixel values of the pixels
horizontally adjacent are 2 and 58, the value added is 60. When
belonging to any one of the monotonous increase/decrease region
2141-1 and monotonous increase/decrease region 2141-2, which are
made up of the pixels in one column vertically arrayed
respectively, and the pixel values of the pixels horizontally
adjacent are 1 and 65, the value added is 66.
It can be understood that the numeric values in one column on the
right side in FIG. 194, i.e., the values obtained by adding the
pixel values on which a fine-line image is projected regarding the
pixels adjacent in the horizontal direction of the two adjacent
monotonous increase/decrease regions made up of the pixels in one
column vertically arrayed, are generally constant.
Similarly, the values obtained by adding the pixel values on which
a fine-line image is projected regarding the pixels adjacent in the
vertical direction of the two adjacent monotonous increase/decrease
regions made up of the pixels in one column horizontally arrayed,
are generally constant.
The allocation-ratio calculation unit 2131 calculates how a
fine-line image is allocated on the pixel values of the pixels in
one column by utilizing characteristics that the values obtained by
adding the pixel values on which the fine-line image is projected
regarding the adjacent pixels of the two adjacent monotonous
increase/decrease regions, are generally constant.
The allocation-ratio calculation unit 2131 calculates, as shown in
FIG. 195, an allocation ratio regarding each pixel belonged to the
two adjacent monotonous increase/decrease regions by dividing the
pixel value of each pixel belonged to the two adjacent monotonous
increase/decrease regions made up of pixels in one column
vertically arrayed by the value obtained by adding the pixel values
on which a fine-line image is projected for each pixel horizontally
adjacent. However, in the event that the calculated result, i.e.,
the calculated allocation ratio exceeds 100, the allocation ratio
is set to 100.
For example, as shown in FIG. 195, when the pixel values of pixels
horizontally adjacent, which are belonged to two adjacent
monotonous increase/decrease regions made up of pixels in one
column vertically arrayed, are 2 and 58 respectively, the value
added is 60, and accordingly, allocation ratios 3.5 and 96.5 are
calculated as to the corresponding pixels respectively. When the
pixel values of pixels horizontally adjacent, which are belonged to
two adjacent monotonous increase/decrease regions made up of pixels
in one column vertically arrayed, are 1 and 65 respectively, the
value added is 65, and accordingly, allocation ratios 1.5 and 98.5
are calculated as to the corresponding pixels respectively.
In this case, in the event that three monotonous increase/decrease
regions are adjacent, regarding which column is first calculated,
of two values obtained by adding the pixel values on which a
fine-line image is projected for each pixel horizontally adjacent,
an allocation ratio is calculated based on a value closer to the
pixel value of the peak P, as shown in FIG. 196.
For example, when the pixel value of the peak P is 81, and the
pixel value of a pixel of interest belonged to a monotonous
increase/decrease region is 79, in the event that the pixel value
of a pixel adjacent to the left side is 3, and the pixel value of a
pixel adjacent to the right side is -1, the value obtained by
adding the pixel value adjacent to the left side is 82, and the
value obtained by adding the pixel value adjacent to the right side
is 78, and consequently, 82 which is closer to the pixel value 81
of the peak P is selected, so an allocation ratio is calculated
based on the pixel adjacent to the left side. Similarly, when the
pixel value of the peak P is 81, and the pixel value of a pixel of
interest belonged to the monotonous increase/decrease region is 75,
in the event that the pixel value of a pixel adjacent to the left
side is 0, and the pixel value of a pixel adjacent to the right
side is 3, the value obtained by adding the pixel value adjacent to
the left side is 75, and the value obtained by adding the pixel
value adjacent to the right side is 78, and consequently, 78 which
is closer to the pixel value 81 of the peak P is selected, so an
allocation ratio is calculated based on the pixel adjacent to the
right side.
Thus, the allocation-ratio calculation unit 2131 calculates an
allocation ratio regarding a monotonous increase/decrease region
made up of pixels in one column vertically arrayed.
With the same processing, the allocation-ratio calculation unit
2131 calculates an allocation ratio regarding a monotonous
increase/decrease region made up of pixels in one column
horizontally arrayed.
The regression-line calculation unit 2132 assumes that the boundary
of a monotonous increase/decrease region is a straight line, and
detects the monotonous increase/decrease region within the
continuity region again by calculating a regression line indicating
the boundary of the monotonous increase/decrease region based on
the calculated allocation ratio by the allocation-ratio calculation
unit 2131.
Description will be made regarding processing for calculating a
regression line indicating the boundary of a monotonous
increase/decrease region in the regression-line calculation unit
2132 with reference to FIG. 197 and FIG. 198.
In FIG. 197, a white circle denotes a pixel positioned in the
boundary on the upper side of the monotonous increase/decrease
region 2141-1 through the monotonous increase/decrease region
2141-5. The regression-line calculation unit 2132 calculates a
regression line regarding the boundary on the upper side of the
monotonous increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 using the regression processing.
For example, the regression-line calculation unit 2132 calculates a
straight line A wherein the sum of squares of the distances with
the pixels positioned in the boundary on the upper side of the
monotonous increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 becomes the minimum value.
Also, in FIG. 197, a black circle denotes a pixel positioned in the
boundary on the lower side of the monotonous increase/decrease
region 2141-1 through the monotonous increase/decrease region
2141-5. The regression-line calculation unit 2132 calculates a
regression line regarding the boundary on the lower side of the
monotonous increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 using the regression processing.
For example, the regression-line calculation unit 2132 calculates a
straight line B wherein the sum of squares of the distances with
the pixels positioned in the boundary on the lower side of the
monotonous increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 becomes the minimum value.
The regression-line calculation unit 2132 detects the monotonous
increase/decrease region within the continuity region again by
determining the boundary of the monotonous increase/decrease region
based on the calculated regression line.
As shown in FIG. 198, the regression-line calculation unit 2132
determines the boundary on the upper side of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 based on the calculated straight
line A. For example, the regression-line calculation unit 2132
determines the boundary on the upper side from the pixel closest to
the calculated straight line A regarding each of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5. For example, the regression-line
calculation unit 2132 determines the boundary on the upper side
such that the pixel closest to the calculated straight line A is
included in each region regarding each of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5.
As shown in FIG. 198, the regression-line calculation unit 2132
determines the boundary on the lower side of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5 based on the calculated straight
line B. For example, the regression-line calculation unit 2132
determines the boundary on the lower side from the pixel closest to
the calculated straight line B regarding each of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5. For example, the regression-line
calculation unit 2132 determines the boundary on the upper side
such that the pixel closest to the calculated straight line B is
included in each region regarding each of the monotonous
increase/decrease region 2141-1 through the monotonous
increase/decrease region 2141-5.
Thus, the regression-line calculation unit 2132 detects a region
wherein the pixel value monotonously increases or decreases from
the peak again based on a regression line for recurring the
boundary of the continuity region detected by the data continuity
detecting unit 101. In other words, the regression-line calculation
unit 2132 detects a region serving as the monotonous
increase/decrease region within the continuity region again by
determining the boundary of the monotonous increase/decrease region
based on the calculated regression line, and supplies region
information indicating the detected region to the line-width
detecting unit 2101.
As described above, the boundary detecting unit 2121 calculates an
allocation ratio indicating proportion wherein a fine-line image
serving as a signal in the actual world 1 is projected on pixels,
and detects the monotonous increase/decrease region within the
continuity region again by calculating a regression line indicating
the boundary of the monotonous increase/decrease region from the
calculated allocation ratio. Thus, a more accurate monotonous
increase/decrease region can be detected.
The line-width detecting unit 2101 shown in FIG. 192 detects the
width of a fine line in the same processing as the case shown in
FIG. 187 based on the region information indicating the region
detected again, which is supplied from the boundary detecting unit
2121. The line-width detecting unit 2101 supplies fine-line width
information indicating the width of a fine line detected to the
signal level estimating unit 2102 along with the data continuity
information.
The processing of the signal level estimating unit 2102 shown in
FIG. 192 is the same processing as the case shown in FIG. 187, so
description thereof will be omitted.
FIG. 199 is a flowchart for describing actual world estimating
processing using the actual world estimating unit 102 of which
configuration is shown in FIG. 192, which corresponds to the
processing in step S102.
In step S2121, the boundary detecting unit 2121 executes boundary
detecting processing for detecting a region again based on the
pixel value of a pixel belonged to the continuity region detected
by the data continuity detecting unit 101. The details of the
boundary detecting processing will be described later.
The processing in step S2122 and step S2123 is the same as the
processing in step S2101 and step S2102, so description thereof
will be omitted.
FIG. 200 is a flowchart for describing boundary detecting
processing corresponding to the processing in step S2121.
In step S2131, the allocation-ratio calculation unit 2131
calculates an allocation ratio indicating proportion wherein a
fine-line image is projected based on the data continuity
information indicating a monotonous increase/decrease region and an
input image. For example, the allocation-ratio calculation unit
2131 detects adjacent monotonous increase/decrease regions within
the continuity region from an input image based on the monotonous
increase/decrease region information included in the data
continuity information, and generates an image made up of
continuity components alone on which a fine-line image is projected
by subtracting approximate values to be approximated at a plane
indicated with a gradient and intercept included in the continuity
component information from the pixel values of the pixels belonged
to the detected monotonous increase/decrease region. Subsequently,
the allocation-ratio calculation unit 2131 calculates an allocation
ratio, by dividing the pixel values of pixels belonged to two
monotonous increase/decrease regions made up of pixels in one
column by the sum of the pixel values of the adjacent pixels,
regarding each pixel belonged to the two adjacent monotonous
increase/decrease regions.
The allocation-ratio calculation unit 2131 supplies the calculated
allocation ratio to the regression-line calculation unit 2132.
In step S2132, the regression-line calculation unit 2132 detects a
region within the continuity region again by calculating a
regression line indicating the boundary of a monotonous
increase/decrease region based on the allocation ratio indicating
proportion wherein a fine-line image is projected. For example, the
regression-line calculation unit 2132 assumes that the boundary of
a monotonous increase/decrease region is a straight line, and
detects the monotonous increase/decrease region within the
continuity region again by calculating a regression line indicating
the boundary of one end of the monotonous increase/decrease region,
and calculating a regression line indicating the boundary of
another end of the monotonous increase/decrease region.
The regression-line calculation unit 2132 supplies region
information indicating the region detected again within the
continuity region to the line-width detecting unit 2101, and the
processing ends.
Thus, the actual world estimating unit 102 of which configuration
is shown in FIG. 192 detects a region made up of pixels on which a
fine-line image is projected again, detects the width of a fine
line in the image serving as a signal in the actual world 1 based
on the region detected again, and estimates the intensity (level)
of light of the signal in the actual world 1. Thus, the width of a
fine line can be detected more accurately, and the intensity of
light can be estimated more accurately regarding a signal in the
actual world 1.
As described above, in the event that a light signal in the real
world is projected, a discontinuous portion of the pixel values of
multiple pixels in the first image data of witch part of continuity
of the light signal in the real world drops is detected, a
continuity region having continuity of data is detected from the
detected discontinuous portion, a region is detected again based on
the pixel values of pixels belonged to the detected continuity
region, and the actual world is estimated based on the region
detected again, a more accurate and higher-precision processing
result can be obtained as to events in the real world.
Next, description will be made regarding the actual world
estimating unit 102 for outputting derivative values of the
approximation function in the spatial direction for each pixel in a
region having continuity as actual world estimating information
with reference to FIG. 201.
A reference-pixel extracting unit 2201 determines regarding whether
or not each pixel in an input image is a processing region based on
the data continuity information (angle as continuity or region
information) input from the data continuity detecting unit 101, and
in the event of a processing region, extracts reference pixel
information necessary for obtaining an approximate function for
approximating the pixel values of pixels in the input image (the
positions and pixel values of multiple pixels around a pixel of
interest necessary for calculation), and outputs this to an
approximation-function estimating unit 2202.
The approximation-function estimating unit 2202 estimates, based on
the least square method, an approximation function for
approximately describing the pixel values of pixels around a pixel
of interest based on the reference pixel information input from the
reference-pixel extracting unit 2201, and outputs the estimated
approximation function to a differential processing unit 2203.
The differential processing unit 2203 obtains a shift amount in the
position of a pixel to be generated from a pixel of interest
according to the angle of the data continuity information (for
example, angle as to a predetermined axis of a fine line or
two-valued edge: gradient) based on the approximation function
input from the approximation-function estimating unit 2202,
calculates a derivative value in the position on the approximation
function according to the shift amount (the derivative value of a
function for approximating the pixel value of each pixel
corresponding to a distance from a line corresponding to continuity
along in the one-dimensional direction), and further, adds
information regarding the position and pixel value of a pixel of
interest, and gradient as continuity to this, and outputs this to
the image generating unit 103 as actual world estimating
information.
Next, description will be made regarding actual world estimating
processing by the actual world estimating unit 102 in FIG. 201 with
reference to the flowchart in FIG. 202.
In step S2201, the reference-pixel extracting unit 2201 acquires an
angle and region information as the data continuity information
from the data continuity detecting unit 101 as well as an input
image.
In step S2202, the reference-pixel extracting unit 2201 sets a
pixel of interest from unprocessed pixels in the input image.
In step S2203, the reference-pixel extracting unit 2201 determines
regarding whether or not the pixel of interest is included in a
processing region based on the region information of the data
continuity information, and in the event that the pixel of interest
is not a pixel in a processing region, the processing proceeds to
step S2210, the differential processing unit 2203 is informed that
the pixel of interest is in a non-processing region via the
approximation-function estimating unit 2202, in response to this,
the differential processing unit 2203 sets the derivative value
regarding the corresponding pixel of interest to zero, further adds
the pixel value of the pixel of interest to this, and outputs this
to the image generating unit 103 as actual world estimating
information, and also the processing proceeds to step S2211. Also,
in the event that determination is made that the pixel of interest
is in a processing region, the processing proceeds to step
S2204.
In step S2204, the reference-pixel extracting unit 2201 determines
regarding whether the direction having data continuity is an angle
close to the horizontal direction or angle close to the vertical
direction based on the angular information included in the data
continuity information. That is to say, in the event that an angle
.theta. having data continuity is
45.degree.>.theta..gtoreq.0.degree., or
180.degree.>.theta..gtoreq.135.degree., the reference-pixel
extracting unit 2201 determines that the direction of continuity of
the pixel of interest is close to the horizontal direction, and in
the event that the angle .theta. having data continuity is
135.degree.>.theta..gtoreq.45.degree., determines that the
direction of continuity of the pixel of interest is close to the
vertical direction.
In step S2205, the reference-pixel extracting unit 2201 extracts
the positional information and pixel values of reference pixels
corresponding to the determined direction from the input image
respectively, and outputs these to the approximation-function
estimating unit 2202. That is to say, reference pixels become data
to be used for calculating a later-described approximation
function, so are preferably extracted according to the gradient
thereof. Accordingly, corresponding to any determined direction of
the horizontal direction and the vertical direction, reference
pixels in a long range in the direction thereof are extracted. More
specifically, for example, as shown in FIG. 203, in the event that
a gradient G.sub.f is close to the vertical direction,
determination is made that the direction is the vertical direction.
In this case, as shown in FIG. 203 for example, when a pixel (0, 0)
in the center of FIG. 203 is taken as a pixel of interest, the
reference-pixel extracting unit 2201 extracts each pixel value of
pixels (-1, 2), (-1, 1), (-1, 0), (-1, -1), (-1, -2), (0, 2), (0,
1), (0, 0), (0, -1), (0, -2), (1, 2), (1, 1), (1, 0), (1, -1), and
(1, -2). Note that in FIG. 203, let us say that both sizes in the
horizontal direction and in the vertical direction of each pixel is
1.
In other words, the reference-pixel extracting unit 2201 extracts
pixels in a long range in the vertical direction as reference
pixels such that the reference pixels are 15 pixels in total of 2
pixels respectively in the vertical (upper/lower) direction.times.1
pixel respectively in the horizontal (left/right) direction
centered on the pixel of interest.
Conversely, in the event that determination is made that the
direction is the horizontal direction, the reference-pixel
extracting unit 2201 extracts pixels in a long range in the
horizontal direction as reference pixels such that the reference
pixels are 15 pixels in total of 1 pixel respectively in the
vertical (upper/lower) direction.times.2 pixels respectively in the
horizontal (left/right) direction centered on the pixel of
interest, and outputs these to the approximation-function
estimating unit 2202. Needless to say, the number of reference
pixels is not restricted to 15 pixels as described above, so any
number of pixels may be employed.
In step S2206, the approximation-function estimating unit 2202
estimates the approximation function f(x) using the least square
method based on information of reference pixels input from the
reference-pixel extracting unit 2201, and outputs this to the
differential processing unit 2203.
That is to say, the approximation function f(x) is a polynomial
such as shown in the following Expression (96).
.function..times..times. ##EQU00062##
Thus, if each of coefficients W.sub.1 through W.sub.n+1 of the
polynomial in Expression (96) can be obtained, the approximation
function f(x) for approximating the pixel value of each reference
pixel (reference pixel value) can be obtained. However, reference
pixel values exceeding the number of coefficients are necessary, so
for example, in the case such as shown in FIG. 203, the number of
reference pixels is 15 pixels in total, and accordingly, the number
of obtainable coefficients in the polynomial is restricted to 15.
In this case, let us say that the polynomial is up to 14-dimension,
and the approximation function is estimated by obtaining the
coefficients W.sub.1 through W.sub.15. Note that in this case,
simultaneous equations may be employed by setting the approximation
function f(x) made up of a 15-dimensional polynomial.
Accordingly, when 15 reference pixel values shown in FIG. 203 are
employed, the approximation-function estimating unit 2202 estimates
the approximation function f(x) by solving the following Expression
(97) using the least square method. P(-1,-2)=f(-1-Cx(-2))
P(-1,-1)=f(-1-Cx(-1)) P(-1,0)=f(-1)(=f(-1-Cx(0)))
P(-1,1)=f(-1-Cx(1)) P(-1,2)=f(-1-Cx(2)) P(0,-2)=f(0-Cx(-2))
P(0,-1)=f(0-Cx(-1)) P(0,0)=f(0)(=f(0-Cx(0))) P(0,1)=f(0-Cx(1))
P(0,2)=f(0-Cx(2)) P(1,-2)=f(1-Cx(-2)) P(1,-1)=f(1-Cx(-1))
P(1,0)=f(1)(=f(1-Cx(0))) P(1,1)=f(1-Cx(1)) P(1,2)=f(1-Cx(2))
(97)
Note that the number of reference pixels may be changed in
accordance with the degree of the polynomial.
Here, Cx (ty) denotes a shift amount, and when the gradient as
continuity is denoted with G.sub.f, Cx(ty)=ty/G.sub.f is defined.
This shift amount Cx (ty) denotes the width of a shift as to the
spatial direction X in the position in the spatial direction Y=ty
on condition that the approximation function f(x) defined on the
position in the spatial direction Y=0 is continuous (has
continuity) along the gradient G.sub.f. Accordingly, for example,
in the event that the approximation function is defined as f (x) on
the position in the spatial direction Y=0, this approximation
function f(x) must be shifted by Cx (ty) as to the spatial
direction X along the gradient G.sub.f in the spatial direction
Y=ty, so the function is defined as
f(x-Cx(ty))(=f(x-ty/G.sub.f).
In step S2207, the differential processing unit 2203 obtains a
shift amount in the position of a pixel to be generated based on
the approximation function f(x) input from the
approximation-function estimating unit 2202.
That is to say, in the event that pixels are generated so as to be
a double density in the horizontal direction and in the vertical
direction respectively (quadruple density in total), the
differential processing unit 2203 first obtains a shift amount of
Pin (Xin, Yin) in the center position to divide a pixel of interest
into two pixels Pa and Pb, which become a double density in the
vertical direction, as shown in FIG. 204, to obtain a derivative
value at a center position Pin (Xin, Yin) of a pixel of interest.
This shift amount becomes Cx (0), so actually becomes zero. Note
that in FIG. 204, a pixel Pin of which general gravity position is
(Xin, Yin) is a square, and pixels Pa and Pb of which general
gravity positions are (Xin, Yin+0.25) and (Xin, Yin-0.25)
respectively are rectangles long in the horizontal direction in the
drawing.
In step S2208, the differential processing unit 2203 differentiates
the approximation function f(x) so as to obtain a primary
differential function f(x)' of the approximation function, obtains
a derivative value at a position according to the obtained shift
amount, and outputs this to the image generating unit 103 as actual
world estimating information. That is to say, in this case, the
differential processing unit 2203 obtains a derivative value f
(Xin)', and adds the position thereof (in this case, a pixel of
interest (Xin, Yin)), the pixel value thereof, and the gradient
information in the direction of continuity to this, and outputs
this.
In step S2209, the differential processing unit 2203 determines
regarding whether or not derivative values necessary for generating
desired-density pixels are obtained. For example, in this case, the
obtained derivative values are only derivative values necessary for
a double density (only derivative values to become a double density
for the spatial direction Y are obtained), so determination is made
that derivative values necessary for generating desired-density
pixels are not obtained, and the processing returns to step
S2207.
In step S2207, the differential processing unit 2203 obtains a
shift amount in the position of a pixel to be generated based on
the approximation function f(x) input from the
approximation-function estimating unit 2202 again. That is to say,
in this case, the differential processing unit 2203 obtains
derivative values necessary for further dividing the divided pixels
Pa and Pb into 2 pixels respectively. The positions of the pixels
Pa and Pb are denoted with black circles in FIG. 204 respectively,
so the differential processing unit 2203 obtains a shift amount
corresponding to each position. The shift amounts of the pixels Pa
and Pb are Cx (0.25) and Cx (-0.25) respectively.
In step S2208, the differential processing unit 2203 subjects the
approximation function f(x) to a primary differentiation, obtains a
derivative value in the position according to a shift amount
corresponding to each of the pixels Pa and Pb, and outputs this to
the image generating unit 103 as actual world estimating
information.
That is to say, in the event of employing the reference pixels
shown in FIG. 203, the differential processing unit 2203, as shown
in FIG. 205, obtains a differential function f(x)' regarding the
obtained approximation function f(x), obtains derivative values in
the positions (Xin-Cx(0.25)) and (Xin-Cx(-0.25)), which are
positions shifted by shift amounts Cx(0.25) and Cx(-0.25) for the
spatial direction X, as f (Xin-Cx(0.25))' and f (Xin-Cx(-0.25))'
respectively, adds the positional information corresponding to the
derivative values thereof to this, and outputs this as actual world
estimating information. Note that the information of the pixel
values is output at the first processing, so this is not added at
this processing.
In step S2209, the differential processing unit 2203 determines
regarding whether or not derivative values necessary for generating
desired-density pixels are obtained again. For example, in this
case, derivative values to become a quadruple density have been
obtained, so determination is made that derivative values necessary
for generating desired-density pixels have been obtained, and the
processing proceeds to step S2211.
In step S2211, the reference-pixel extracting unit 2201 determines
regarding whether or not all of the pixels have been processed, and
in the event that determination is made that all of the pixels have
not been processed, the processing returns to step S2202. Also, in
step S2211, in the event that determination is made that all of the
pixels have been processed, the processing ends.
As described above, in the event that pixels are generated so as to
become a quadruple density in the horizontal direction and in the
vertical direction regarding the input image, pixels are divided by
extrapolation/interpolation using the derivative value of the
approximation function in the center position of the pixel to be
divided, so in order to generate quadruple-density pixels,
information of three derivative values in total is necessary.
That is to say, as shown in FIG. 204, derivative values necessary
for generating four pixels P01, P02, P03, and P04 (in FIG. 204,
pixels P01, P02, P03, and P04 are squares of which the gravity
positions are the positions of four cross marks in the drawing, and
the length of each side is 1 for the pixel Pin, so around 0.5 for
the pixels P01, P02, P03, and P04) are necessary for one pixel in
the end, and accordingly, in order to generate quadruple-density
pixels, first, double-density pixels in the horizontal direction or
in the vertical direction (in this case, in the vertical direction)
are generated (the above first processing in steps S2207 and
S2208), and further, the divided two pixels are divided in the
direction orthogonal to the initial dividing direction (in this
case, in the horizontal direction) (the above second processing in
steps S2207 and S2208).
Note that with the above example, description has been made
regarding derivative values at the time of calculating
quadruple-density pixels as an example, but in the event of
calculating pixels having a density more than a quadruple density,
many more derivative values necessary for calculating pixel values
may be obtained by repeatedly performing the processing in steps
S2207 through S2209. Also, with the above example, description has
been made regarding an example for obtaining double-density pixel
values, but the approximation function f(x) is a continuous
function, so necessary derivative values may be obtained even
regarding pixel values having a density other than a pluralized
density.
According to the above arrangement, an approximation function for
approximately expressing the pixel values of pixels near a pixel of
interest can be obtained, and derivative values in the positions
corresponding to the pixel positions in the spatial direction can
be output as actual world estimating information.
With the actual world estimating unit 102 described in FIG. 201,
derivative values necessary for generating an image have been
output as actual world estimating information, but a derivative
value is the same value as a gradient of the approximation function
f(x) in a necessary position.
Now, description will be made next regarding the actual world
estimating unit 102 wherein gradients alone on the approximation
function f(x) necessary for generating pixels are directly obtained
without obtaining the approximation function f(x), and output as
actual world estimating information, with reference to FIG.
206.
The reference-pixel extracting unit 2211 determines regarding
whether or not each pixel of an input image is a processing region
based on the data continuity information (angle as continuity, or
region information) input from the data continuity detecting unit
101, and in the event of a processing region, extracts information
of reference pixels necessary for obtaining gradients from the
input image (perimeter multiple pixels arrayed in the vertical
direction including a pixel of interest, which are necessary for
calculation, or the positions of perimeter multiple pixels arrayed
in the horizontal direction including a pixel of interest, and
information of each pixel value), and outputs this to a gradient
estimating unit 2212.
The gradient estimating unit 2212 generates gradient information of
a pixel position necessary for generating a pixel based on the
reference pixel information input from the reference-pixel
extracting unit 2211, and outputs this to the image generating unit
103 as actual world estimating information. More specifically, the
gradient estimating unit 2212 obtains a gradient in the position of
a pixel of interest on the approximation function f(x)
approximately expressing the actual world using the difference
information of the pixel values between pixels, outputs this along
with the position information and pixel value of the pixel of
interest, and the gradient information in the direction of
continuity, as actual world estimating information.
Next, description will be made regarding the actual world
estimating processing by the actual world estimating unit 102 in
FIG. 206 with reference to the flowchart in FIG. 207.
In step S2221, the reference-pixel extracting unit 2211 acquires an
angle and region information as the data continuity information
from the data continuity detecting unit 101 along with an input
image.
In step S2222, the reference-pixel extracting unit 2211 sets a
pixel of interest from unprocessed pixels in the input image.
In step S2223, the reference-pixel extracting unit 2211 determines
regarding whether or not the pixel of interest is in a processing
region based on the region information of the data continuity
information, and in the event that determination is made that the
pixel of interest is not a pixel in the processing region, the
processing proceeds to step S2228, wherein the gradient estimating
unit 2212 is informed that the pixel of interest is in a
non-processing region, in response to this, the gradient estimating
unit 2212 sets the gradient for the corresponding pixel of interest
to zero, and further adds the pixel value of the pixel of interest
to this, and outputs this as actual world estimating information to
the image generating unit 103, and also the processing proceeds to
step S2229. Also, in the event that determination is made that the
pixel of interest is in a processing region, the processing
proceeds to step S2224.
In step S2224, the reference-pixel extracting unit 2211 determines
regarding whether the direction having data continuity is an angle
close to the horizontal direction or angle close to the vertical
direction based on the angular information included in the data
continuity information. That is to say, in the event that an angle
.theta. having data continuity is
45.degree.>.theta..gtoreq.0.degree., or
180.degree.>.theta..gtoreq.135.degree., the reference-pixel
extracting unit 2211 determines that the direction of continuity of
the pixel of interest is close to the horizontal direction, and in
the event that the angle .theta. having data continuity is
135.degree.>.theta..gtoreq.45.degree., determines that the
direction of continuity of the pixel of interest is close to the
vertical direction.
In step S2225, the reference-pixel extracting unit 2211 extracts
the positional information and pixel values of reference pixels
corresponding to the determined direction from the input image
respectively, and outputs these to the gradient estimating unit
2212. That is to say, reference pixels become data to be used for
calculating a later-described gradient, so are preferably extracted
according to a gradient indicating the direction of continuity.
Accordingly, corresponding to any determined direction of the
horizontal direction and the vertical direction, reference pixels
in a long range in the direction thereof are extracted. More
specifically, for example, in the event that determination is made
that a gradient is close to the vertical direction, as shown in
FIG. 208, when a pixel (0, 0) in the center of FIG. 208 is taken as
a pixel of interest, the reference-pixel extracting unit 2211
extracts each pixel value of pixels (0, 2), (0, 1), (0, 0), (0,
-1), and (0, -2). Note that in FIG. 208, let us say that both sizes
in the horizontal direction and in the vertical direction of each
pixel is 1.
In other words, the reference-pixel extracting unit 2211 extracts
pixels in a long range in the vertical direction as reference
pixels such that the reference pixels are 5 pixels in total of 2
pixels respectively in the vertical (upper/lower) direction
centered on the pixel of interest.
Conversely, in the event that determination is made that the
direction is the horizontal direction, the reference-pixel
extracting unit 2211 extracts pixels in a long range in the
horizontal direction as reference pixels such that the reference
pixels are 5 pixels in total of 2 pixels respectively in the
horizontal (left/right) direction centered on the pixel of
interest, and outputs these to the approximation-function
estimating unit 2202. Needless to say, the number of reference
pixels is not restricted to 5 pixels as described above, so any
number of pixels may be employed.
In step S2226, the gradient estimating unit 2212 calculates a shift
amount of each pixel value based on the reference pixel information
input from the reference-pixel extracting unit 2211, and the
gradient G.sub.f in the direction of continuity. That is to say, in
the event that the approximation function f(x) corresponding to the
spatial direction Y=0 is taken as a basis, the approximation
functions corresponding to the spatial directions Y=-2, -1, 1, and
2 are continuous along the gradient G.sub.f as continuity as shown
in FIG. 208, so the respective approximation functions are
described as f (x-Cx(2)), f (x-Cx(1)), f (x-Cx(-1)), and f
(x-Cx(-2)), and are represented as functions shifted by each shift
amount in the spatial direction X for each of the spatial
directions Y=-2, -1, 1, and 2.
Accordingly, the gradient estimating unit 2212 obtains shift
amounts Cx (-2) through Cx (2) of these. For example, in the event
that reference pixels are extracted such as shown in FIG. 208, with
regard to the shift amounts thereof, the reference pixel (0, 2) in
the drawing becomes Cx (2)=2/G.sub.f, the reference pixel (0, 1)
becomes Cx (1)=1/G.sub.f, the reference pixel (0, 0) becomes Cx
(0)=0, the reference pixel (0, -1) becomes Cx (-1)=-1/G.sub.f, and
the reference pixel (0, -2) becomes Cx (-2)=-2/G.sub.f.
In step S2227, the gradient estimating unit 2212 calculates
(estimates) a gradient on the approximation function f(x) in the
position of the pixel of interest. For example, as shown in FIG.
208, in the event that the direction of continuity regarding the
pixel of interest is an angle close to the vertical direction, the
pixel values between the pixels adjacent in the horizontal
direction exhibit great differences, but change between the pixels
in the vertical direction is small and similar, and accordingly,
the gradient estimating unit 2212 substitutes the difference
between the pixels in the vertical direction for the difference
between the pixels in the horizontal direction, and obtains a
gradient on the approximation function f(x) in the position of the
pixel of interest, by seizing change between the pixels in the
vertical direction as change in the spatial direction X according
to a shift amount.
That is to say, if we assume that the approximation function f(x)
approximately describing the real world exists, the relations
between the above shift amounts and the pixel values of the
respective reference pixels is such as shown in FIG. 209. Here, the
pixel values of the respective pixels in FIG. 208 are represented
as P (0, 2), P (0, 1), P (0, 0), P (0, -1), and P (0, -2) from the
top. As a result, with regard to the pixel value P and shift amount
Cx near the pixel of interest (0, 0), 5 pairs of relations (P,
Cx)=((P (0, 2), -Cx (2)), (P (0, 1), -Cx (1)), (P (0, -1)), -Cx
(-1)), (P (0, -2), -Cx (-2)), and (P (0, 0), 0) are obtained.
Now, with the pixel value P, shift amount Cx, and gradient Kx
(gradient on the approximation function f(x)), the relation such as
the following Expression (98) holds. P=Kx.times.Cx (98)
The above Expression (98) is a one-variable function regarding the
variable Kx, so the gradient estimating unit 2212 obtains the
gradient Kx (gradient) using the least square method of one
variable.
That is to say, the gradient estimating unit 2212 obtains the
gradient of the pixel of interest by solving a normal equation such
as shown in the following Expression (99), adds the pixel value of
the pixel of interest, and the gradient information in the
direction of continuity to this, and outputs this to the image
generating unit 103 as actual world estimating information.
.times..times. ##EQU00063##
Here, i denotes a number for identifying each pair of the pixel
value P and shift amount C of the above reference pixel, 1 through
m. Also, m denotes the number of the reference pixels including the
pixel of interest.
In step S2229, the reference-pixel extracting unit 2211 determines
regarding whether or not all of the pixels have been processed, and
in the event that determination is made that all of the pixels have
not been processed, the processing returns to step S2222. Also, in
the event that determination is made that all of the pixels have
been processed in step S2229, the processing ends.
Note that the gradient to be output as actual world estimating
information by the above processing is employed at the time of
calculating desired pixel values to be obtained finally through
extrapolation/interpolation. Also, with the above example,
description has been made regarding the gradient at the time of
calculating double-density pixels as an example, but in the event
of calculating pixels having a density more than a double density,
gradients in many more positions necessary for calculating the
pixel values may be obtained.
For example, as shown in FIG. 204, in the event that pixels having
a quadruple density in the spatial directions in total of a double
density in the horizontal direction and also a double density in
the vertical direction are generated, the gradient Kx of the
approximation function f(x) corresponding to the respective
positions Pin, Pa, and Pb in FIG. 204 should be obtained, as
described above.
Also, with the above example, an example for obtaining
double-density pixels has been described, but the approximation
function f(x) is a continuous function, so it is possible to obtain
a necessary gradient even regarding the pixel value of a pixel in a
position other than a pluralized density.
According to the above arrangements, it is possible to generate and
output gradients on the approximation function necessary for
generating pixels in the spatial direction as actual world
estimating information by using the pixel values of pixels near a
pixel of interest without obtaining the approximation function
approximately representing the actual world.
Next, description will be made regarding the actual world
estimating unit 102, which outputs derivative values on the
approximation function in the frame direction (temporal direction)
for each pixel in a region having continuity as actual world
estimating information, with reference to FIG. 210.
The reference-pixel extracting unit 2231 determines regarding
whether or not each pixel in an input image is in a processing
region based on the data continuity information (movement as
continuity (movement vector), and region information) input from
the data continuity detecting unit 101, and in the event that each
pixel is in a processing region, extracts reference pixel
information necessary for obtaining an approximation function
approximating the pixel values of the pixels in the input image
(multiple pixel positions around a pixel of interest necessary for
calculation, and the pixel values thereof), and outputs this to the
approximation-function estimating unit 2202.
The approximation-function estimating unit 2232 estimates an
approximation function, which approximately describes the pixel
value of each pixel around the pixel of interest based on the
reference pixel information in the frame direction input from the
reference-pixel extracting unit 2231, based on the least square
method, and outputs the estimated function to the differential
processing unit 2233.
The differential processing unit 2233 obtains a shift amount in the
frame direction in the position of a pixel to be generated from the
pixel of interest according to the movement of the data continuity
information based on the approximation function in the frame
direction input from the approximation-function estimating unit
2232, calculates a derivative value in a position on the
approximation function in the frame direction according to the
shift amount thereof (derivative value of the function
approximating the pixel value of each pixel corresponding to a
distance along in the primary direction from a line corresponding
to continuity), further adds the position and pixel value of the
pixel of interest, and information regarding movement as continuity
to this, and outputs this to the image generating unit 103 as
actual world estimating information.
Next, description will be made regarding the actual world
estimating processing by the actual world estimating unit 102 in
FIG. 210 with reference to the flowchart in FIG. 211.
In step S2241, the reference-pixel extracting unit 2231 acquires
the movement and region information as the data continuity
information from the data continuity detecting unit 101 along with
an input image.
In step S2242, the reference-pixel extracting unit 2231 sets a
pixel of interest from unprocessed pixels in the input image.
In step S2243, the reference-pixel extracting unit 2231 determines
regarding whether or not the pixel of interest is included in a
processing region based on the region information of the data
continuity information, and in the event that the pixel of interest
is not a pixel in a processing region, the processing proceeds to
step S2250, the differential processing unit 2233 is informed that
the pixel of interest is in a non-processing region via the
approximation-function estimating unit 2232, in response to this,
the differential processing unit 2233 sets the derivative value
regarding the corresponding pixel of interest to zero, further adds
the pixel value of the pixel of interest to this, and outputs this
to the image generating unit 103 as actual world estimating
information, and also the processing proceeds to step S2251. Also,
in the event that determination is made that the pixel of interest
is in a processing region, the processing proceeds to step
S2244.
In step S2244, the reference-pixel extracting unit 2231 determines
regarding whether the direction having data continuity is movement
close to the spatial direction or movement close to the frame
direction based on movement information included in the data
continuity information. That is to say, as shown in FIG. 212, if we
say that an angle indicating the spatial and temporal directions
within a surface made up of the frame direction T, which is taken
as a reference axis, and the spatial direction Y, is taken as
.theta.v, in the event that an angle .theta.v having data
continuity is 45.degree.>.theta.v.gtoreq.0.degree., or
180.degree.>.theta.v.gtoreq.135.degree., the reference-pixel
extracting unit 2201 determines that the movement as continuity of
the pixel of interest is close to the frame direction (temporal
direction), and in the event that the angle .theta. having data
continuity is 135.degree.>.theta..gtoreq.45.degree., determines
that the direction of continuity of the pixel of interest is close
to the spatial direction.
In step S2245, the reference-pixel extracting unit 2201 extracts
the positional information and pixel values of reference pixels
corresponding to the determined direction from the input image
respectively, and outputs these to the approximation-function
estimating unit 2232. That is to say, reference pixels become data
to be used for calculating a later-described approximation
function, so are preferably extracted according to the angle
thereof. Accordingly, corresponding to any determined direction of
the frame direction and the spatial direction, reference pixels in
a long range in the direction thereof are extracted. More
specifically, for example, as shown in FIG. 212, in the event that
a movement direction V.sub.f is close to the spatial direction,
determination is made that the direction is the spatial direction.
In this case, as shown in FIG. 212 for example, when a pixel (t,
y)=(0, 0) in the center of FIG. 212 is taken as a pixel of
interest, the reference-pixel extracting unit 2231 extracts each
pixel value of pixels (t, y)=(-1, 2), (-1, 1), (-1, 0), (-1, -1),
(-1, -2), (0, 2), (0, 1), (0, 0), (0, -1), (0, -2), (1, 2), (1, 1),
(1, 0), (1, -1), and (1, -2). Note that in FIG. 212, let us say
that both sizes in the frame direction and in the spatial direction
of each pixel is 1.
In other words, the reference-pixel extracting unit 2231 extracts
pixels in a long range in the spatial direction as to the frame
direction as reference pixels such that the reference pixels are 15
pixels in total of 2 pixels respectively in the spatial direction
(upper/lower direction in the drawing).times.1 frame respectively
in the frame direction (left/right direction in the drawing)
centered on the pixel of interest.
Conversely, in the event that determination is made that the
direction is the frame direction, the reference-pixel extracting
unit 2231 extracts pixels in a long range in the frame direction as
reference pixels such that the reference pixels are 15 pixels in
total of 1 pixel respectively in the spatial direction (upper/lower
direction in the drawing).times.2 frames respectively in the frame
direction (left/right direction in the drawing) centered on the
pixel of interest, and outputs these to the approximation-function
estimating unit 2232. Needless to say, the number of reference
pixels is not restricted to 15 pixels as described above, so any
number of pixels may be employed.
In step S2246, the approximation-function estimating unit 2232
estimates the approximation function f(t) using the least square
method based on information of reference pixels input from the
reference-pixel extracting unit 2231, and outputs this to the
differential processing unit 2233.
That is to say, the approximation function f(t) is a polynomial
such as shown in the following Expression (100).
.function..times..times. ##EQU00064##
Thus, if each of coefficients W.sub.1 through W.sub.n+1 of the
polynomial in Expression (100) can be obtained, the approximation
function f(t) in the frame direction for approximating the pixel
value of each reference pixel can be obtained. However, reference
pixel values exceeding the number of coefficients are necessary, so
for example, in the case such as shown in FIG. 212, the number of
reference pixels is 15 pixels in total, and accordingly, the number
of obtainable coefficients in the polynomial is restricted to 15.
In this case, let us say that the polynomial is up to 14-dimension,
and the approximation function is estimated by obtaining the
coefficients W.sub.1 through W.sub.15. Note that in this case,
simultaneous equations may be employed by setting the approximation
function f(t) made up of a 15-dimensional polynomial.
Accordingly, when 15 reference pixel values shown in FIG. 212 are
employed, the approximation-function estimating unit 2232 estimates
the approximation function f(t) by solving the following Expression
(101) using the least square method. P(-1,-2)=f(-1-Ct(-2))
P(-1,-1)=f(-1-Ct(-1)) P(-1,0)=f(-1)(=f(-1-Ct(0)))
P(-1,1)=f(-1-Ct(1)) P(-1,2)=f(-1-Ct(2)) P(0,-2)=f(0-Ct(-2))
P(0,-1)=f(0-Ct(-1)) P(0,0)=f(0)(=f(0-Ct(0))) P(0,1)=f(0-Ct(1))
P(0,2)=f(0-Ct(2)) P(1,-2)=f(1-Ct(-2)) P(1,-1)=f(1-Ct(-1))
P(1,0)=f(1)(=f(1-Ct(0))) P(1,1)=f(1-Ct(1)) P(1,2)=f(1-Ct(2))
(101)
Note that the number of reference pixels may be changed in
accordance with the degree of the polynomial.
Here, Ct (ty) denotes a shift amount, which is the same as the
above Cx (ty), and when the gradient as continuity is denoted with
V.sub.f, Ct (ty)=ty/V.sub.f is defined. This shift amount Ct (ty)
denotes the width of a shift as to the frame direction T in the
position in the spatial direction Y=ty on condition that the
approximation function f(t) defined on the position in the spatial
direction Y=0 is continuous (has continuity) along the gradient
V.sub.f. Accordingly, for example, in the event that the
approximation function is defined as f (t) on the position in the
spatial direction Y=0, this approximation function f(t) must be
shifted by Ct (ty) as to the frame direction (temporal direction) T
in the spatial direction Y=ty, so the function is defined as f
(t-Ct (ty)) (=f (t-ty/V.sub.f).
In step S2247, the differential processing unit 2233 obtains a
shift amount in the position of a pixel to be generated based on
the approximation function f(t) input from the
approximation-function estimating unit 2232.
That is to say, in the event that pixels are generated so as to be
a double density in the frame direction and in the spatial
direction respectively (quadruple density in total), the
differential processing unit 2233 first obtains, for example, a
shift amount of later-described Pin (Tin, Yin) in the center
position to be divided into later-described two pixels Pat and Pbt,
which become a double density in the spatial direction, as shown in
FIG. 213, to obtain a derivative value at a center position Pin
(Tin, Yin) of a pixel of interest. This shift amount becomes Ct
(0), so actually becomes zero. Note that in FIG. 213, a pixel Pin
of which general gravity position is (Tin, Yin) is a square, and
pixels Pat and Pbt of which general gravity positions are (Tin,
Yin+0.25) and (Tin, Yin-0.25) respectively are rectangles long in
the horizontal direction in the drawing. Also, the length in the
frame direction T of the pixel of interest Pin is 1, which
corresponds to the shutter time for one frame.
In step S2248, the differential processing unit 2233 differentiates
the approximation function f(t) so as to obtain a primary
differential function f(t)' of the approximation function, obtains
a derivative value at a position according to the obtained shift
amount, and outputs this to the image generating unit 103 as actual
world estimating information. That is to say, in this case, the
differential processing unit 2233 obtains a derivative value f
(Tin)', and adds the position thereof (in this case, a pixel of
interest (Tin, Yin)), the pixel value thereof, and the movement
information in the direction of continuity to this, and outputs
this.
In step S2249, the differential processing unit 2233 determines
regarding whether or not derivative values necessary for generating
desired-density pixels are obtained. For example, in this case, the
obtained derivative values are only derivative values necessary for
a double density in the spatial direction (derivative values to
become a double density for the frame direction are not obtained),
so determination is made that derivative values necessary for
generating desired-density pixels are not obtained, and the
processing returns to step S2247.
In step S2247, the differential processing unit 2203 obtains a
shift amount in the position of a pixel to be generated based on
the approximation function f(t) input from the
approximation-function estimating unit 2202 again. That is to say,
in this case, the differential processing unit 2203 obtains
derivative values necessary for further dividing the divided pixels
Pat and Pbt into 2 pixels respectively. The positions of the pixels
Pat and Pbt are denoted with black circles in FIG. 213
respectively, so the differential processing unit 2233 obtains a
shift amount corresponding to each position. The shift amounts of
the pixels Pat and Pbt are Ct (0.25) and Ct (-0.25)
respectively.
In step S2248, the differential processing unit 2233 differentiates
the approximation function f(t), obtains a derivative value in the
position according to a shift amount corresponding to each of the
pixels Pat and Pbt, and outputs this to the image generating unit
103 as actual world estimating information.
That is to say, in the event of employing the reference pixels
shown in FIG. 212, the differential processing unit 2233, as shown
in FIG. 214, obtains a differential function f(t)' regarding the
obtained approximation function f(t), obtains derivative values in
the positions (Tin-Ct (0.25)) and (Tin-Ct (-0.25)), which are
positions shifted by shift amounts Ct (0.25) and Ct (-0.25) for the
spatial direction T, as f (Tin-Ct (0.25))' and f (Tin-Ct (-0.25))'
respectively, adds the positional information corresponding to the
derivative values thereof to this, and outputs this as actual world
estimating information. Note that the information of the pixel
values is output at the first processing, so this is not added at
this processing.
In step S2249, the differential processing unit 2233 determines
regarding whether or not derivative values necessary for generating
desired-density pixels are obtained again. For example, in this
case, derivative values to become a double density in the spatial
direction Y and in the frame direction T respectively (quadruple
density in total) are obtained, so determination is made that
derivative values necessary for generating desired-density pixels
are obtained, and the processing proceeds to step S2251.
In step S2251, the reference-pixel extracting unit 2231 determines
regarding whether or not all of the pixels have been processed, and
in the event that determination is made that all of the pixels have
not been processed, the processing returns to step S2242. Also, in
step S2251, in the event that determination is made that all of the
pixels have been processed, the processing ends.
As described above, in the event that pixels are generated so as to
become a quadruple density in the frame direction (temporal
direction) and in the spatial direction regarding the input image,
pixels are divided by extrapolation/interpolation using the
derivative value of the approximation function in the center
position of the pixel to be divided, so in order to generate
quadruple-density pixels, information of three derivative values in
total is necessary.
That is to say, as shown in FIG. 213, derivative values necessary
for generating four pixels P01t, P02t, P03t, and P04t (in FIG. 213,
pixels P01t, P02t, P03t, and P04t are squares of which the gravity
positions are the positions of four cross marks in the drawing, and
the length of each side is 1 for the pixel Pin, so around 0.5 for
the pixels P01t, P02t, P03t, and P04t) are necessary for one pixel
in the end, and accordingly, in order to generate quadruple-density
pixels, first, double-density pixels in the frame direction or in
the spatial direction are generated (the above first processing in
steps S2247 and S2248), and further, the divided two pixels are
divided in the direction orthogonal to the initial dividing
direction (in this case, in the frame direction) (the above second
processing in steps S2247 and S2248).
Note that with the above example, description has been made
regarding derivative values at the time of calculating
quadruple-density pixels as an example, but in the event of
calculating pixels having a density more than a quadruple density,
many more derivative values necessary for calculating pixel values
may be obtained by repeatedly performing the processing in steps
S2247 through S2249. Also, with the above example, description has
been made regarding an example for obtaining double-density pixel
values, but the approximation function f(t) is a continuous
function, so derivative values may be obtained even regarding pixel
values having a density other than a pluralized density.
According to the above arrangement, an approximation function for
approximately expressing the pixel value of each pixel near a pixel
of interest can be obtained, and derivative values in the positions
necessary for generating pixels can be output as actual world
estimating information.
With the actual world estimating unit 102 described in FIG. 210,
derivative values necessary for generating an image have been
output as actual world estimating information, but a derivative
value is the same value as a gradient of the approximation function
f(t) in a necessary position.
Now, description will be made next regarding the actual world
estimating unit 102 wherein gradients alone in the frame direction
on the approximation function necessary for generating pixels are
directly obtained without obtaining the approximation function, and
output as actual world estimating information, with reference to
FIG. 215.
A reference-pixel extracting unit 2251 determines regarding whether
or not each pixel of an input image is a processing region based on
the data continuity information (movement as continuity, or region
information) input from the data continuity detecting unit 101, and
in the event of a processing region, extracts information of
reference pixels necessary for obtaining gradients from the input
image (perimeter multiple pixels arrayed in the spatial direction
including a pixel of interest, which are necessary for calculation,
or the positions of perimeter multiple pixels arrayed in the frame
direction including a pixel of interest, and information of each
pixel value), and outputs this to a gradient estimating unit
2252.
The gradient estimating unit 2252 generates gradient information of
a pixel position necessary for generating a pixel based on the
reference pixel information input from the reference-pixel
extracting unit 2251, and outputs this to the image generating unit
103 as actual world estimating information. In further detail the
gradient estimating unit 2252 obtains a gradient in the frame
direction in the position of a pixel of interest on the
approximation function approximately expressing the pixel value of
each reference pixel using the difference information of the pixel
values between pixels, outputs this along with the position
information and pixel value of the pixel of interest, and the
movement information in the direction of continuity, as actual
world estimating information.
Next, description will be made regarding the actual world
estimating processing by the actual world estimating unit 102 in
FIG. 215 with reference to the flowchart in FIG. 216.
In step S2261, the reference-pixel extracting unit 2251 acquires
movement and region information as the data continuity information
from the data continuity detecting unit 101 along with an input
image.
In step S2262, the reference-pixel extracting unit 2251 sets a
pixel of interest from unprocessed pixels in the input image.
In step S2263, the reference-pixel extracting unit 2251 determines
regarding whether or not the pixel of interest is in a processing
region based on the region information of the data continuity
information, and in the event that determination is made that the
pixel of interest is not a pixel in a processing region, the
processing proceeds to step S2268, wherein the gradient estimating
unit 2252 is informed that the pixel of interest is in a
non-processing region, in response to this, the gradient estimating
unit 2252 sets the gradient for the corresponding pixel of interest
to zero, and further adds the pixel value of the pixel of interest
to this, and outputs this as actual world estimating information to
the image generating unit 103, and also the processing proceeds to
step S2269. Also, in the event that determination is made that the
pixel of interest is in a processing region, the processing
proceeds to step S2264.
In step S2264, the reference-pixel extracting unit 2211 determines
regarding whether movement as data continuity is movement close to
the frame direction or movement close to the spatial direction
based on the movement information included in the data continuity
information. That is to say, if we say that an angle indicating the
spatial and temporal directions within a surface made up of the
frame direction T, which is taken as a reference axis, and the
spatial direction Y, is taken as .theta.v, in the event that an
angle .theta.v of movement as data continuity is
45.degree.>.theta.v.gtoreq.0.degree., or
180.degree.>.theta.v.gtoreq.135.degree., the reference-pixel
extracting unit 2251 determines that the movement as continuity of
the pixel of interest is close to the frame direction, and in the
event that the angle .theta.v having data continuity is
135.degree.>.theta.v.gtoreq.45.degree., determines that the
movement as continuity of the pixel of interest is close to the
spatial direction.
In step S2265, the reference-pixel extracting unit 2251 extracts
the positional information and pixel values of reference pixels
corresponding to the determined direction from the input image
respectively, and outputs these to the gradient estimating unit
2252. That is to say, reference pixels become data to be used for
calculating a later-described gradient, so are preferably extracted
according to movement as continuity. Accordingly, corresponding to
any determined direction of the frame direction and the spatial
direction, reference pixels in a long range in the direction
thereof are extracted. More specifically, for example, in the event
that determination is made that movement is close to the spatial
direction, as shown in FIG. 217, when a pixel (t, y)=(0, 0) in the
center of FIG. 217 is taken as a pixel of interest, the
reference-pixel extracting unit 2251 extracts each pixel value of
pixels (t, y)=(0, 2), (0, 1), (0, 0), (0, -1), and (0, -2). Note
that in FIG. 217, let us say that both sizes in the frame direction
and in the spatial direction of each pixel is 1.
In other words, the reference-pixel extracting unit 2251 extracts
pixels in a long range in the spatial direction as reference pixels
such that the reference pixels are 5 pixels in total of 2 pixels
respectively in the spatial direction (upper/lower direction in the
drawing) centered on the pixel of interest.
Conversely, in the event that determination is made that the
direction is the frame direction, the reference-pixel extracting
unit 2251 extracts pixels in a long range in the horizontal
direction as reference pixels such that the reference pixels are 5
pixels in total of 2 pixels respectively in the frame direction
(left/right direction in the drawing) centered on the pixel of
interest, and outputs these to the approximation-function
estimating unit 2252. Needless to say, the number of reference
pixels is not restricted to 5 pixels as described above, so any
number of pixels may be employed.
In step S2266, the gradient estimating unit 2252 calculates a shift
amount of each pixel value based on the reference pixel information
input from the reference-pixel extracting unit 2251, and the
movement V.sub.f in the direction of continuity. That is to say, in
the event that the approximation function f(t) corresponding to the
spatial direction Y=0 is taken as a basis, the approximation
functions corresponding to the spatial directions Y=-2, -1, 1, and
2 are continuous along the gradient V.sub.f as continuity as shown
in FIG. 217, so the respective approximation functions are
described as f (t-Ct (2)), f (t-Ct (1)), f (t-Ct (-1)), and f (t-Ct
(-2)), and are represented as functions shifted by each shift
amount in the frame direction T for each of the spatial directions
Y=-2, -1, 1, and 2.
Accordingly, the gradient estimating unit 2252 obtains shift
amounts Ct (-2) through Ct (2) of these. For example, in the event
that reference pixels are extracted such as shown in FIG. 217, with
regard to the shift amounts thereof, the reference pixel (0, 2) in
the drawing becomes Ct (2)=2/V.sub.f, the reference pixel (0, 1)
becomes Ct (1)=1/V.sub.f, the reference pixel (0, 0) becomes Ct
(0)=0, the reference pixel (0, -1) becomes Ct (-1)=-1/V.sub.f, and
the reference pixel (0, -2) becomes Ct (-2)=-2/V.sub.f. The
gradient estimating unit 2252 obtains these shift amounts Ct (-2)
through Ct (2).
In step S2267, the gradient estimating unit 2252 calculates
(estimates) a gradient in the frame direction of the pixel of
interest. For example, as shown in FIG. 217, in the event that the
direction of continuity regarding the pixel of interest is an angle
close to the spatial direction, the pixel values between the pixels
adjacent in the frame direction exhibit great differences, but
change between the pixels in the spatial direction is small and
similar, and accordingly, the gradient estimating unit 2252
substitutes the difference between the pixels in the frame
direction for the difference between the pixels in the spatial
direction, and obtains a gradient at the pixel of interest, by
seizing change between the pixels in the spatial direction as
change in the frame direction T according to a shift amount.
That is to say, if we assume that the approximation function f(t)
approximately describing the real world exists, the relations
between the above shift amounts and the pixel values of the
respective reference pixels is such as shown in FIG. 218. Here, the
pixel values of the respective pixels in FIG. 218 are represented
as P (0, 2), P (0, 1), P (0, 0), P (0, -1), and P (0, -2) from the
top. As a result, with regard to the pixel value P and shift amount
Ct near the pixel of interest (0, 0), 5 pairs of relations (P,
Ct)=((P (0, 2), -Ct (2)), (P (0, 1), -Ct (1)), (P (0, -1)), -Ct
(-1)), (P (0, -2), -Ct (-2)), and (P (0, 0), 0) are obtained.
Now, with the pixel value P, shift amount Ct, and gradient Kt
(gradient on the approximation function f(t)), the relation such as
the following Expression (102) holds. P=Kt.times.Ct (102)
The above Expression (102) is a one-variable function regarding the
variable Kt, so the gradient estimating unit 2212 obtains the
variable Kt (gradient) using the least square method of one
variable.
That is to say, the gradient estimating unit 2252 obtains the
gradient of the pixel of interest by solving a normal equation such
as shown in the following Expression (103), adds the pixel value of
the pixel of interest, and the gradient information in the
direction of continuity to this, and outputs this to the image
generating unit 103 as actual world estimating information.
.times..times. ##EQU00065##
Here, i denotes a number for identifying each pair of the pixel
value P and shift amount Ct of the above reference pixel, 1 through
m. Also, m denotes the number of the reference pixels including the
pixel of interest.
In step S2269, the reference-pixel extracting unit 2251 determines
regarding whether or not all of the pixels have been processed, and
in the event that determination is made that all of the pixels have
not been processed, the processing returns to step S2262. Also, in
the event that determination is made that all of the pixels have
been processed in step S2269, the processing ends.
Note that the gradient in the frame direction to be output as
actual world estimating information by the above processing is
employed at the time of calculating desired pixel values to be
obtained finally through extrapolation/interpolation. Also, with
the above example, description has been made regarding the gradient
at the time of calculating double-density pixels as an example, but
in the event of calculating pixels having a density more than a
double density, gradients in many more positions necessary for
calculating the pixel values may be obtained.
For example, as shown in FIG. 204, in the event that pixels having
a quadruple density in the temporal and spatial directions in total
of a double density in the horizontal direction and also a double
density in the frame direction are generated, the gradient Kt of
the approximation function f(t) corresponding to the respective
positions Pin, Pat, and Pbt in FIG. 204 should be obtained, as
described above.
Also, with the above example, an example for obtaining
double-density pixel values has been described, but the
approximation function f(t) is a continuous function, so it is
possible to obtain a necessary gradient even regarding the pixel
value of a pixel in a position other than a pluralized density.
Needless to say, there is no restriction regarding the sequence of
processing for obtaining gradients on the approximation function as
to the frame direction or the spatial direction or derivative
values. Further, with the above example in the spatial direction,
description has been made using the relation between the spatial
direction Y and frame direction T, but the relation between the
spatial direction X and frame direction T may be employed instead
of this. Further, a gradient (in any one-dimensional direction) or
a derivative value may be selectively obtained from any
two-dimensional relation of the temporal and spatial
directions.
According to the above arrangements, it is possible to generate and
output gradients on the approximation function in the frame
direction (temporal direction) of positions necessary for
generating pixels as actual world estimating information by using
the pixel values of pixels near a pixel of interest without
obtaining the approximation function in the frame direction
approximately representing the actual world.
Next, description will be made regarding another embodiment example
of the actual world estimating unit 102 (FIG. 3) with reference to
FIG. 219 through FIG. 249.
FIG. 219 is a diagram for describing the principle of this
embodiment example.
As shown in FIG. 219, a signal (light intensity allocation) in the
actual world 1, which is an image cast on the sensor 2, is
represented with a predetermined function F. Note that hereafter,
with the description of this embodiment example, the signal serving
as an image in the actual world 1 is particularly referred to as a
light signal, and the function F is particularly referred to as a
light signal function F.
With this embodiment example, in the event that the light signal in
the actual world 1 represented with the light signal function F has
predetermined continuity, the actual world estimating unit 102
estimates the light signal function F by approximating the light
signal function F with a predetermined function f using an input
image (image data including continuity of data corresponding to
continuity) from the sensor 2, and data continuity information
(data continuity information corresponding to continuity of the
input image data) from the data continuity detecting unit 101. Note
that with the description of this embodiment example, the function
f is particularly referred to as an approximation function f,
hereafter.
In other words, with this embodiment example, the actual world
estimating unit 102 approximates (describes) the image (light
signal in the actual world 1) represented with the light signal
function F using a model 161 (FIG. 7) represented with the
approximation function f. Accordingly, hereafter, this embodiment
example is referred to as a function approximating method.
Now, description will be made regarding the background wherein the
present applicant has invented the function approximating method,
prior to entering the specific description of the function
approximating method.
FIG. 220 is a diagram for describing integration effects in the
case in which the sensor 2 is treated as a CCD.
As shown in FIG. 220, multiple detecting elements 2-1 are disposed
on the plane of the sensor 2.
With the example in FIG. 220, a direction in parallel with a
predetermined side of the detecting elements 2-1 is taken as the X
direction, which is one direction in the spatial direction, and the
a direction orthogonal to the X direction is taken as the Y
direction, which is another direction in the spatial direction.
Also, the direction perpendicular to the X-Y plane is taken as the
direction t serving as the temporal direction.
Also, with the example in FIG. 220, the spatial shape of each
detecting element 2-1 of the sensor 2 is represented with a square
of which one side is 1 in length. The shutter time (exposure time)
of the sensor 2 is represented with 1.
Further, with the example in FIG. 220, the center of one detecting
element 2-1 of the sensor 2 is taken as the origin (position x=0 in
the X direction, and position y=0 in the Y direction) in the
spatial direction (X direction and Y direction), and the
intermediate point-in-time of the exposure time is taken as the
origin (position t=0 in the t direction) in the temporal direction
(t direction).
In this case, the detecting element 2-1 of which the center is in
the origin (x=0, y=0) in the spatial direction subjects the light
signal function F(x, y, t) to integration with a range between -0.5
and 0.5 in the X direction, range between -0.5 and 0.5 in the Y
direction, and range between -0.5 and 0.5 in the t direction, and
outputs the integral value thereof as a pixel value P.
That is to say, the pixel value P output from the detecting element
2-1 of which the center is in the origin in the spatial direction
is represented with the following Expression (104).
.intg..times..intg..times..intg..times..function..times.d.times.d.times.d
##EQU00066##
The other detecting elements 2-1 also output the pixel value P
shown in Expression (104) by taking the center thereof as the
origin in the spatial direction in the same way.
FIG. 221 is a diagram for describing a specific example of the
integration effects of the sensor 2.
In FIG. 221, the X direction and Y direction represent the X
direction and Y direction of the sensor 2 (FIG. 220).
A portion 2301 of the light signal in the actual world 1
(hereafter, such a portion is referred to as a region) represents
an example of a region having predetermined continuity.
Note that the region 2301 is a portion of the continuous light
signal (continuous region). On the other hand, in FIG. 221, the
region 2301 is shown as divided into 20 small regions (square
regions) in reality. This is because of representing that the size
of the region 2301 is equivalent to the size wherein the four
detecting elements (pixels) of the sensor 2 in the X direction, and
also the five detecting elements (pixels) of the sensor 2 in the Y
direction are arrayed. That is to say, each of the 20 small regions
(virtual regions) within the region 2301 is equivalent to one
pixel.
Also, a white portion within the region 2301 represents a light
signal corresponding to a fine line. Accordingly, the region 2301
has continuity in the direction wherein a fine line continues.
Hereafter, the region 2301 is referred to as the
fine-line-including actual world region 2301.
In this case, when the fine-line-including actual world region 2301
(a portion of a light signal in the actual world 1) is detected by
the sensor 2, region 2302 (hereafter, this is referred to as a
fine-line-including data region 2302) of the input image (pixel
values) is output from the sensor 2 by integration effects.
Note that each pixel of the fine-line-including data region 2302 is
represented as an image in the drawing, but is data representing a
predetermined value in reality. That is to say, the
fine-line-including actual world region 2301 is changed (distorted)
to the fine-line-including data region 2302, which is divided into
20 pixels (20 pixels in total of 4 pixels in the X direction and
also 5 pixels in the Y direction) each having a predetermined pixel
value by the integration effects of the sensor 2.
FIG. 222 is a diagram for describing another specific example
(example different from FIG. 221) of the integration effects of the
sensor 2.
In FIG. 222, the X direction and Y direction represent the X
direction and Y direction of the sensor 2 (FIG. 220).
A portion (region) 2303 of the light signal in the actual world 1
represents another example (example different from the
fine-line-including actual region 2301 in FIG. 221) of a region
having predetermined continuity.
Note that the region 2303 is a region having the same size as the
fine-line-including actual world region 2301. That is to say, the
region 2303 is also a portion of the continuous light signal in the
actual world 1 (continuous region) as with the fine-line-including
actual world region 2301 in reality, but is shown as divided into
20 small regions (square regions) equivalent to one pixel of the
sensor 2 in FIG. 222.
Also, the region 2303 includes a first portion edge having
predetermined first light intensity (value), and a second portion
edge having predetermined second light intensity (value).
Accordingly, the region 2303 has continuity in the direction
wherein the edges continue. Hereafter, the region 2303 is referred
to as the two-valued-edge-including actual world region 2303.
In this case, when the two-valued-edge-including actual world
region 2303 (a portion of the light signal in the actual world 1)
is detected by the sensor 2, a region 2304 (hereafter, referred to
as two-valued-edge-including data region 2304) of the input image
(pixel value) is output from the sensor 2 by integration
effects.
Note that each pixel value of the two-valued-edge-including data
region 2304 is represented as an image in the drawing as with the
fine-line-including data region 2302, but is data representing a
predetermined value in reality. That is to say, the
two-valued-edge-including actual world region 2303 is changed
(distorted) to the two-valued-edge-including data region 2304,
which is divided into 20 pixels (20 pixels in total of 4 pixels in
the X direction and also 5 pixels in the Y direction) each having a
predetermined pixel value by the integration effects of the sensor
2.
Conventional image processing devices have regarded image data
output from the sensor 2 such as the fine-line-including data
region 2302, two-valued-edge-including data region 2304, and the
like as the origin (basis), and also have subjected the image data
to the subsequent image processing. That is to say, regardless of
that the image data output from the sensor 2 had been changed
(distorted) to data different from the light signal in the actual
world 1 by integration effects, the conventional image processing
devices have performed image processing on assumption that the data
different from the light signal in the actual world 1 is
correct.
As a result, the conventional image processing devices have
provided a problem wherein based on the waveform (image data) of
which the details in the actual world is distorted at the stage
wherein the image data is output from the sensor 2, it is very
difficult to restore the original details from the waveform.
Accordingly, with the function approximating method, in order to
solve this problem, as described above (as shown in FIG. 219), the
actual world estimating unit 102 estimates the light signal
function F by approximating the light signal function F(light
signal in the actual world 1) with the approximation function f
based on the image data (input image) such as the
fine-line-including data region 2302, and two-valued-edge-including
data region 2304 output from the sensor 2.
Thus, at a later stage than the actual world estimating unit 102
(in this case, the image generating unit 103 in FIG. 3), the
processing can be performed by taking the image data wherein
integration effects are taken into consideration, i.e., image data
that can be represented with the approximation function f as the
origin.
Hereafter, description will be made independently regarding three
specific methods (first through third function approximating
methods), of such a function approximating method with reference to
the drawings.
First, description will be made regarding the first function
approximating method with reference to FIG. 223 through FIG.
237.
FIG. 223 is a diagram representing the fine-line-including actual
world region 2301 shown in FIG. 221 described above again.
In FIG. 223, the X direction and Y direction represent the X
direction and Y direction of the sensor 2 (FIG. 220).
The first function approximating method is a method for
approximating a one-dimensional waveform (hereafter, such a
waveform is referred to as an X cross-sectional waveform F(x))
wherein the light signal function F(x, y, t) corresponding to the
fine-line-including actual world region 2301 such as shown in FIG.
223 is projected in the X direction (direction of an arrow 2311 in
the drawing), with the approximation function f(x) serving as an
n-dimensional (n is an arbitrary integer) polynomial. Accordingly,
hereafter, the first function approximating method is particularly
referred to as a one-dimensional polynomial approximating
method.
Note that with the one-dimensional polynomial approximating method,
the X cross-sectional waveform F(x), which is to be approximated,
is not restricted to a waveform corresponding to the
fine-line-including actual world region 2301 in FIG. 223, of
course. That is to say, as described later, with the
one-dimensional polynomial approximating method, any waveform can
be approximated as long as the X cross-sectional waveform F(x)
corresponds to the light signals in the actual world 1 having
continuity.
Also, the direction of the projection of the light signal function
F(x, y, t) is not restricted to the X direction, or rather the Y
direction or t direction may be employed. That is to say, with the
one-dimensional polynomial approximating method, a function F(y)
wherein the light signal function F(x, y, t) is projected in the Y
direction may be approximated with a predetermined approximation
function f(y), or a function F(t) wherein the light signal function
F(x, y, t) is projected in the t direction may be approximated with
a predetermined approximation f (t).
More specifically, the one-dimensional polynomial approximating
method is a method for approximating, for example, the X
cross-sectional waveform F(x) with the approximation function f(x)
serving as an n-dimensional polynomial such as shown in the
following Expression (105).
.function..times..times..times..times..times. ##EQU00067##
That is to say, with the one-dimensional polynomial approximating
method, the actual world estimating unit 102 estimates the X
cross-sectional waveform F(x) by calculating the coefficient
(features) w.sub.i of x^i in Expression (105).
This calculation method of the features w.sub.i is not restricted
to a particular method, for example, the following first through
third methods may be employed.
That is to say, the first method is a method that has been employed
so far.
On the other hand, the second method is a method that has been
newly invented by the present applicant, which is a method that
considers continuity in the spatial direction as to the first
method.
However, as described later, with the first and second methods, the
integration effects of the sensor 2 are not taken into
consideration. Accordingly, an approximation function f(x) obtained
by substituting the features w.sub.i calculated by the first method
or the second method for the above Expression (105) is an
approximation function regarding an input image, but strictly
speaking, cannot be referred to as the approximation function of
the X cross-sectional waveform F(x).
Consequently, the present applicant has invented the third method
that calculates the features w.sub.i further in light of the
integration effects of the sensor 2 as to the second method. An
approximation function f(x) obtained by substituting the features
w.sub.i calculated with this third method for the above Expression
(105) can be referred to as the approximation function of the X
cross-sectional waveform F(x) in that the integration effects of
the sensor 2 are taken into consideration.
Thus, strictly speaking, the first method and the second method
cannot be referred to as the one-dimensional polynomial
approximating method, and the third method alone can be referred to
as the one-dimensional polynomial approximating method.
In other words, as shown in FIG. 224, the second method is an
embodiment of the actual world estimating unit 102 according to the
present invention, which is different from the one-dimensional
polynomial approximating method. That is to say, FIG. 224 is a
diagram for describing the principle of the embodiment
corresponding to the second method.
As shown in FIG. 224, with the embodiment corresponding to the
second method, in the event that the light signal in the actual
world 1 represented with the light signal function F has
predetermined continuity, the actual world estimating unit 102 does
not approximate the X cross-sectional waveform F(x) with an input
image (image data including continuity of data corresponding to
continuity) from the sensor 2, and data continuity information
(data continuity information corresponding to continuity of input
image data) from the data continuity detecting unit 101, but
approximates the input image from the sensor 2 with a predetermined
approximation function f.sub.2 (x).
Thus, it is hard to say that the second method is a method having
the same level as the third method in that approximation of the
input image alone is performed without considering the integral
effects of the sensor 2. However, the second method is a method
superior to the conventional first method in that the second method
takes continuity in the spatial direction into consideration.
Hereafter, description will be made independently regarding the
details of the first method, second method, and third method in
this order.
Note that hereafter, in the event that the respective approximation
functions f (x) generated by the first method, second method, and
third method are distinguished from that of the other method, they
are particularly referred to as approximation function f.sub.1 (x),
approximation function f.sub.2 (x), and approximation function
f.sub.3 (x) respectively.
First, description will be made regarding the details of the first
method.
With the first method, on condition that the approximation function
f.sub.1 (x) shown in the above Expression (105) holds within the
fine-line-including actual world region 2301 in FIG. 225, the
following prediction equation (106) is defined.
.function..function. ##EQU00068##
In Expression (106), x represents a pixel position relative as to
the X direction from a pixel of interest. y represents a pixel
position relative as to the Y direction from the pixel of interest.
e represents a margin of error. Specifically, for example, as shown
in FIG. 225, let us say that the pixel of interest is the second
pixel in the X direction from the left, and also the third pixel in
the Y direction from the bottom in the drawing, of the
fine-line-including data region 2302 (data of which the
fine-line-including actual world region 2301 (FIG. 223) is detected
by the sensor 2, and output). Also, let us say that the center of
the pixel of interest is the origin (0, 0), and a coordinates
system (hereafter, referred to as a pixel-of-interest coordinates
system) of which axes are an x axis and y axis respectively in
parallel with the X direction and Y direction of the sensor 2 (FIG.
220) is set. In this case, the coordinates value (x, y) of the
pixel-of-interest coordinates system represents a relative pixel
position.
Also, in Expression (106), P (x, y) represents a pixel value in the
relative pixel positions (x, y). Specifically, in this case, the P
(x, y) within the fine-line-including data region 2302 is such as
shown in FIG. 226.
FIG. 226 represents this pixel value P (x, y) in a graphic
manner.
In FIG. 226, the respective vertical axes of the graphs represent
pixel values, and the horizontal axes represent a relative position
x in the X direction from the pixel of interest. Also, in the
drawing, the dashed line in the first graph from the top represents
an input pixel value P (x, -2), the broken triple-dashed line in
the second graph from the top represents an input pixel value P (x,
-1), the solid line in the third graph from the top represents an
input pixel value P (x, 0), the broken line in the fourth graph
from the top represents an input pixel value P (x, 1), and the
broken double-dashed line in the fifth graph from the top (the
first from the bottom) represents an input pixel value P (x, 2)
respectively.
Upon the 20 input pixel values P (x, -2), P (x, -1), P (x, 0), P
(x, 1), and P (x, 2) (however, x is any one integer value of -1
through 2) shown in FIG. 226 being substituted for the above
Expression (106) respectively, 20 equations as shown in the
following Expression (107) are generated. Note that each e.sub.k (k
is any one of integer values 1 through 20) represents a margin of
error.
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function..times..times..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..times..t-
imes..function..function..times..times..function..function..times..times..-
function..function..times..times..function..function..times..times..functi-
on..function..times..times..function..function..times..times..function..fu-
nction..times..times..function..function..times..times..function..function-
. ##EQU00069##
Expression (107) is made up of 20 equations, so in the event that
the number of the features w.sub.i of the approximation function
f.sub.1 (x) is less than 20, i.e., in the event that the
approximation function f.sub.1 (x) is a polynomial having the
number of dimensions less than 19, the features w.sub.i can be
calculated using the least square method, for example. Note that
the specific solution of the least square method will be described
later.
For example, if we say that the number of dimensions of the
approximation function f.sub.1 (x) is five, the approximation
function f.sub.1 (x) calculated with the least square method using
Expression (107) (the approximation function f.sub.1 (x) generated
by the calculated features w.sub.i) becomes a curve shown in FIG.
227.
Note that in FIG. 227, the vertical axis represents pixel values,
and the horizontal axis represents a relative position x from the
pixel of interest.
That is to say, for example, if we supplement the respective 20
pixel values P (x, y) (the respective input pixel values P (x, -2),
P (x, -1), P (x, 0), P (x, 1), and P (x, 2) shown in FIG. 226)
making up the fine-line-including data region 2302 in FIG. 225
along the x axis without any modification (if we regard a relative
position y in the Y direction as constant, and overlay the five
graphs shown in FIG. 226), multiple lines (dashed line, broken
triple-dashed line, solid line, broken line, and broken
double-dashed line) in parallel with the x axis, such as shown in
FIG. 227, are distributed.
However, in FIG. 227, the dashed line represents the input pixel
value P (x, -2), the broken triple-dashed line represents the input
pixel value P (x, -1), the solid line represents the input pixel
value P (x, 0), the broken line represents the input pixel value P
(x, 1), and the broken double-dashed line represents the input
pixel value P (x, 2) respectively. Also, in the event of the same
pixel value, lines more than 2 lines are overlaid in reality, but
in FIG. 227, the lines are drawn so as to distinguish each line,
and so as not to overlay each line.
The respective 20 input pixel values (P (x, -2), P (x, -1), P (x,
0), P (x, 1), and P (x, 2)) thus distributed, and a regression
curve (the approximation function f.sub.1 (x) obtained by
substituting the features w.sub.i calculated with the least square
method for the above Expression (104)) so as to minimize the error
of the value f.sub.1 (x) become a curve (approximation function
f.sub.1 (x)) shown in FIG. 227.
Thus, the approximation function f.sub.1 (x) represents nothing but
a curve connecting in the X direction the means of the pixel values
(pixel values having the same relative position x in the X
direction from the pixel of interest) P (x, -2), P (x, -1), P (x,
0), P (x, 1), and P (x, 2) in the Y direction. That is to say, the
approximation function f.sub.1 (x) is generated without considering
continuity in the spatial direction included in the light
signal.
For example, in this case, the fine-line-including actual world
region 2301 (FIG. 223) is regarded as a subject to be approximated.
This fine-line-including actual world region 2301 has continuity in
the spatial direction, which is represented with a gradient
G.sub.F, such as shown in FIG. 228. Note that in FIG. 228, the X
direction and Y direction represent the X direction and Y direction
of the sensor 2 (FIG. 220).
Accordingly, the data continuity detecting unit 101 (FIG. 219) can
output an angle .theta. (angle .theta. generated between the
direction of data continuity represented with a gradient G.sub.f
corresponding to the gradient G.sub.F, and the X direction) such as
shown in FIG. 228 as data continuity information corresponding to
the gradient G.sub.F as continuity in the spatial direction.
However, with the first method, the data continuity information
output from the data continuity detecting unit 101 is not employed
at all.
In other words, such as shown in FIG. 228, the direction of
continuity in the spatial direction of the fine-line-including
actual world region 2301 is a general angle .theta. direction.
However, the first method is a method for calculating the features
w.sub.i of the approximation function f.sub.1 (x) on assumption
that the direction of continuity in the spatial direction of the
fine-line-including actual world region 2301 is the Y direction
(i.e., on assumption that the angle .theta. is 90.degree.).
Consequently, the approximation function f.sub.1 (x) becomes a
function of which the waveform gets dull, and the detail decreases
than the original pixel value. In other words, though not shown in
the drawing, with the approximation function f.sub.1 (x) generated
with the first method, the waveform thereof becomes a waveform
different from the actual X cross-sectional waveform F(x).
To this end, the present applicant has invented the second method
for calculating the features w.sub.i by further taking continuity
in the spatial direction into consideration (utilizing the angle
.theta.) as to the first method.
That is to say, the second method is a method for calculating the
features w.sub.i of the approximation function f.sub.2 (x) on
assumption that the direction of continuity of the
fine-line-including actual world region 2301 is a general angle
.theta. direction.
Specifically, for example, the gradient G.sub.f representing
continuity of data corresponding to continuity in the spatial
direction is represented with the following Expression (108).
.times..times..theta.dd ##EQU00070##
Note that in Expression (108), dx represents the amount of fine
movement in the X direction such as shown in FIG. 228, dy
represents the amount of fine movement in the Y direction as to the
dx such as shown in FIG. 228.
In this case, if we define the shift amount C.sub.x (y) as shown in
the following Expression (109), with the second method, an equation
corresponding to Expression (106) employed in the first method
becomes such as the following Expression (110).
.function..function..function..function. ##EQU00071##
That is to say, Expression (106) employed in the first method
represents that the position x in the X direction of the pixel
center position (x, y) is the same value regarding the pixel value
P (x, y) of any pixel positioned in the same position. In other
words, Expression (106) represents that pixels having the same
pixel value continue in the Y direction (exhibits continuity in the
Y direction).
On the other hand, Expression (110) employed in the second method
represents that the pixel value P (x, y) of a pixel of which the
center position is (x, y) is not identical to the pixel value
(approximate equivalent to f.sub.2 (x)) of a pixel positioned in a
place distant from the pixel of interest (a pixel of which the
center position is the origin (0, 0)) in the X direction by x, and
is the same value as the pixel value (approximate equivalent to
f.sub.2 (x+C.sub.x (y)) of a pixel positioned in a place further
distant from the pixel thereof in the X direction by the shift
amount C.sub.x (y) (pixel positioned in a place distant from the
pixel of interest in the X direction by x+C.sub.x (y)). In other
words, Expression (110) represents that pixels having the same
pixel value continue in the angle .theta. direction corresponding
to the shift amount C.sub.x (y) (exhibits continuity in the general
angle .theta. direction).
Thus, the shift amount C.sub.x (y) is the amount of correction
considering continuity (in this case, continuity represented with
the gradient G.sub.F in FIG. 228 (strictly speaking, continuity of
data represented with the gradient G.sub.f)) in the spatial
direction, and Expression (110) is obtained by correcting
Expression (106) with the shift amount C.sub.x (y).
In this case, upon the 20 pixel values P (x, y) (however, x is any
one integer value of -1 through 2, and y is any one integer value
of -2 through 2) of the fine-line-including data region shown in
FIG. 225 being substituted for the above Expression (110)
respectively, 20 equations as shown in the following Expression
(111) are generated.
.function..function..function..times..times..function..function..function-
..times..times..function..function..function..times..times..function..func-
tion..function..times..times..function..function..function..times..times..-
function..function..function..times..times..function..function..function..-
times..times..function..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..function-
..times..times..function..function..function..times..times..function..func-
tion..function..times..times..function..function..function..times..times..-
function..function..function..times..times..function..function..function..-
times..times..function..function..function..times..times..function..functi-
on..function. ##EQU00072##
Expression (111) is made up of 20 equations, as with the above
Expression (107). Accordingly, with the second method, as with the
first method, in the event that the number of the features w.sub.i
of the approximation function f.sub.2 (x) is less than 20, i.e.,
the approximation function f.sub.2 (x) is a polynomial having the
number of dimensions less than 19, the features w.sub.i can be
calculated with the least square method, for example. Note that the
specific solution regarding the least square method will be
described later.
For example, if we say that the number of dimensions of the
approximation function f.sub.2 (x) is five as with the first
method, with the second method, the features w.sub.i are calculated
as follows.
That is to say, FIG. 229 represents the pixel value P (x, y) shown
in the left side of Expression (111) in a graphic manner. The
respective five graphs shown in FIG. 229 are basically the same as
shown in FIG. 226.
As shown in FIG. 229, the maximal pixel values (pixel values
corresponding to fine lines) are continuous in the direction of
continuity of data represented with the gradient G.sub.f.
Consequently, with the second method, if we supplement the
respective input pixel values P (x, -2), P (x, -1), P (x, 0), P (x,
1), and P (x, 2) shown in FIG. 229, for example, along the x axis,
we supplement the pixel values after the pixel values are changed
in the states shown in FIG. 230 instead of supplementing the pixel
values without any modification as with the first method (let us
assume that y is constant, and the five graphs are overlaid in the
states shown in FIG. 229).
That is to say, FIG. 230 represents a state wherein the respective
input pixel values P (x, -2), P (x, -1), P (x, 0), P (x, 1), and P
(x, 2) shown in FIG. 229 are shifted by the shift amount C.sub.x
(y) shown in the above Expression (109). In other words, FIG. 230
represents a state wherein the five graphs shown in FIG. 229 are
moved as if the gradient G.sub.F representing the actual direction
of continuity of data were regarded as a gradient G.sub.F' (in the
drawing, a straight line made up of a dashed line were regarded as
a straight line made up of a solid line).
In the states in FIG. 230, if we supplement the respective input
pixel values P (x, -2), P (x, -1), P (x, 0), P (x, 1), and P (x,
2), for example, along the x axis (in the states shown in FIG. 230,
if we overlay the five graphs), multiple lines (dashed line, broken
triple-dashed line, solid line, broken line, and broken
double-dashed line) in parallel with the x axis, such as shown in
FIG. 231, are distributed.
Note that in FIG. 231, the vertical axis represents pixel values,
and the horizontal axis represents a relative position x from the
pixel of interest. Also, the dashed line represents the input pixel
value P (x, -2), the broken triple-dashed line represents the input
pixel value P (x, -1), the solid line represents the input pixel
value P (x, 0), the broken line represents the input pixel value P
(x, 1), and the broken double-dashed line represents the input
pixel value P (x, 2) respectively. Further, in the event of the
same pixel value, lines more than 2 lines are overlaid in reality,
but in FIG. 231, the lines are drawn so as to distinguish each
line, and so as not to overlay each line.
The respective 20 input pixel values P (x, y) (however, x is any
one integer value of -1 through 2, and y is any one integer value
of -2 through 2) thus distributed, and a regression curve (the
approximation function f.sub.2 (x) obtained by substituting the
features w.sub.i calculated with the least square method for the
above Expression (104)) to minimize the error of the value f.sub.2
(x+C.sub.x (y)) become a curve f.sub.2 (x) shown in the solid line
in FIG. 231.
Thus, the approximation function f.sub.2 (x) generated with the
second method represents a curve connecting in the X direction the
means of the input pixel values P (x, y) in the angle .theta.
direction (i.e., direction of continuity in the general spatial
direction) output from the data continuity detecting unit 101 (FIG.
219).
On the other hand, as described above, the approximation function
f.sub.1 (x) generated with the first method represents nothing but
a curve connecting in the X direction the means of the input pixel
values P (x, y) in the Y direction (i.e., the direction different
from the continuity in the spatial direction).
Accordingly, as shown in FIG. 231, the approximation function
f.sub.2 (x) generated with the second method becomes a function
wherein the degree of dullness of the waveform thereof decreases,
and also the degree of decrease of the detail as to the original
pixel value decreases less than the approximation function f.sub.1
(x) generated with the first method. In other words, though not
shown in the drawing, with the approximation function f.sub.2 (x)
generated with the second method, the waveform thereof becomes a
waveform closer to the actual X cross-sectional waveform F(x) than
the approximation function f.sub.1 (x) generated with the first
method.
However, as described above, the approximation function f.sub.2 (x)
is a function considering continuity in the spatial direction, but
is nothing but a function generated wherein the input image (input
pixel value) is regarded as the origin (basis). That is to say, as
shown in FIG. 224 described above, the approximation function
f.sub.2 (x) is nothing but a function that approximated the input
image different from the X cross-sectional waveform F(x), and it is
hard to say that the approximation function f.sub.2 (x) is a
function that approximated the X cross-sectional waveform F(x). In
other words, the second method is a method for calculating the
features w.sub.i on assumption that the above Expression (110)
holds, but does not take the relation in Expression (104) described
above into consideration (does not consider the integration effects
of the sensor 2).
Consequently, the present applicant has invented the third method
that calculates the features w.sub.i of the approximation function
f.sub.3 (x) by further taking the integration effects of the sensor
2 into consideration as to the second method.
That is to say the third method is a method that introduces the
concept of a spatial mixed region.
Description will be made regarding a spatial mixed region with
reference to FIG. 232 prior to description of the third method.
In FIG. 232, a portion 2321 (hereafter, referred to as a region
2321) of a light signal in the actual world 1 represents a region
having the same area as one detecting element (pixel) of the sensor
2.
Upon the sensor 2 detecting the region 2321, the sensor 2 outputs a
value (one pixel value) 2322 obtained by the region 2321 being
subjected to integration in the temporal and spatial directions (X
direction, Y direction, and t direction). Note that the pixel value
2322 is represented as an image in the drawing, but is actually
data representing a predetermined value.
The region 2321 in the actual world 1 is clearly classified into a
light signal (white region in the drawing) corresponding to the
foreground (the above fine line, for example), and a light signal
(black region in the drawing) corresponding to the background.
On the other hand, the pixel value 2322 is a value obtained by the
light signal in the actual world 1 corresponding to the foreground
and the light signal in the actual world 1 corresponding to the
background being subjected to integration. In other words, the
pixel value 2322 is a value corresponding to a level wherein the
light corresponding to the foreground and the light corresponding
to the background are spatially mixed.
Thus, in the event that a portion corresponding to one pixel
(detecting element of the sensor 2) of the light signals in the
actual world 1 is not a portion where the light signals having the
same level are spatially uniformly distributed, but a portion where
the light signals having a different level such as a foreground and
background are distributed, upon the region thereof being detected
by the sensor 2, the region becomes one pixel value as if the
different light levels were spatially mixed by the integration
effects of the sensor 2 (integrated in the spatial direction).
Thus, a region made up of pixels in which an image (light signals
in the actual world 1) corresponding to a foreground, and an image
(light signals in the actual world 1) corresponding to a background
are subjected to spatial integration is, here, referred to as a
spatial mixed region.
Accordingly, with the third method, the actual world estimating
unit 102 (FIG. 219) estimates the X cross-sectional waveform F(x)
representing the original region 2321 in the actual world 1 (of the
light signals in the actual world 1, the portion 2321 corresponding
to one pixel of the sensor 2) by approximating the X
cross-sectional waveform F(x) with the approximation function
f.sub.3 (x) serving as a one-dimensional polynomial such as shown
in FIG. 233.
That is to say, FIG. 233 represents an example of the approximation
function f.sub.3 (x) corresponding to the pixel value 2322 serving
as a spatial mixed region (FIG. 232), i.e., the approximation
function f.sub.3 (x) that approximates the X cross-sectional
waveform F(x) corresponding to the solid line within the region
2331 in the actual world 1 (FIG. 232). In FIG. 233, the axis in the
horizontal direction in the drawing represents an axis in parallel
with the side from the upper left end x.sub.s to lower right end
x.sub.e of the pixel corresponding to the pixel value 2322 (FIG.
232), which is taken as the x axis. The axis in the vertical
direction in the drawing is taken as an axis representing pixel
values.
In FIG. 233, the following Expression (112) is defined on condition
that the result obtained by subjecting the approximation function
f.sub.3 (x) to integration in a range (pixel width) from the
x.sub.s to the x.sub.e is generally identical with the pixel values
P (x, y) output from the sensor 2 (dependent on a margin of error e
alone).
.times..intg..times..function..times.d.times..intg..times..times..times..-
times..times.d.times..function..times..times. ##EQU00073##
In this case, the features w.sub.i of the approximation function
f.sub.3 (x) are calculated from the 20 pixel values P (x, y)
(however, x is any one integer value of -1 through 2, and y is any
one integer value of -2 through 2) of the fine-line-including data
region 2302 shown in FIG. 228, so the pixel value P in Expression
(112) becomes the pixel values P (x, y).
Also, as with the second method, it is necessary to take continuity
in the spatial direction into consideration, and accordingly, each
of the start position x.sub.s and end position x.sub.e in the
integral range in Expression (112) is dependent upon the shift
amount C.sub.x (y). That is to say, each of the start position
x.sub.s and end position x.sub.e of the integral range in
Expression (112) is represented such as the following Expression
(113).
.function..times..times..function. ##EQU00074##
In this case, upon each pixel value of the fine-line-including data
region 2302 shown in FIG. 228, i.e., each of the input pixel values
P (x, -2), P (x, -1), P (x, 0), P (x, 1), and P (x, 2) (however, x
is any one integer value of -1 through 2) shown in FIG. 229 being
substituted for the above Expression (112) (the integral range is
the above Expression (113)), 20 equations shown in the following
Expression (114) are generated.
.function..intg..function..function..times..function..times.d.times..func-
tion..intg..function..function..times..function..times.d.times..function..-
intg..function..function..times..function..times.d.times..function..intg..-
function..function..times..function..times.d.times..function..intg..functi-
on..function..times..function..times.d.times..function..intg..function..fu-
nction..times..function..times.d.times..function..intg..function..function-
..times..function..times.d.times..function..intg..function..function..time-
s..function..times.d.times..function..intg..times..function..times.d.times-
..function..intg..times..function..times.d.times..function..intg..times..f-
unction..times.d.times..function..intg..times..function..times.d.times..fu-
nction..intg..function..function..times..function..times.d.times..function-
..intg..function..function..times..function..times.d.times..function..intg-
..function..function..times..function..times.d.times..function..intg..func-
tion..function..times..function..times.d.times..function..intg..function..-
function..times..function..times.d.times..function..intg..function..functi-
on..times..function..times.d.times..function..intg..function..function..ti-
mes..function..times.d.times..function..intg..function..function..times..f-
unction..times.d ##EQU00075##
Expression (114) is made up of 20 equations as with the above
Expression (111). Accordingly, with the third method as with the
second method, in the event that the number of the features w.sub.i
of the approximation function f.sub.3 (x) is less than 20, i.e., in
the event that the approximation function f.sub.3 (x) is a
polynomial having the number of dimensions less than 19, for
example, the features w.sub.i may be calculated with the least
square method. Note that the specific solution of the least square
method will be described later.
For example, if we say that the number of dimensions of the
approximation function f.sub.3 (x) is five, the approximation
function f.sub.3 (x) calculated with the least square method using
Expression (114) (the approximation function f.sub.3 (x) generated
with the calculated features w.sub.i) becomes a curve shown with
the solid line in FIG. 234.
Note that in FIG. 234, the vertical axis represents pixel values,
and the horizontal axis represents a relative position x from the
pixel of interest.
As shown in FIG. 234, in the event that the approximation function
f.sub.3 (x) (a curve shown with a solid line in the drawing)
generated with the third method is compared with the approximation
function f.sub.2 (x) (a curve shown with a dashed line in the
drawing) generated with the second method, a pixel value at x=0
becomes great, and also the gradient of the curve creates a steep
waveform. This is because details increase more than the input
pixels, resulting in being unrelated to the resolution of the input
pixels. That is to say, we can say that the approximation function
f.sub.3 (x) approximates the X cross-sectional waveform F(x).
Accordingly, though not shown in the drawing, the approximation
function f.sub.3 (x) becomes a waveform closer to the X
cross-sectional waveform F(x) than the approximation function
f.sub.2 (x).
FIG. 235 represents an configuration example of the actual world
estimating unit 102 employing such a one-dimensional polynomial
approximating method.
In FIG. 235, the actual world estimating unit 102 estimates the X
cross-sectional waveform F(x) by calculating the features w.sub.i
using the above third method (least square method), and generating
the approximation function f(x) of the above Expression (105) using
the calculated features w.sub.i.
As shown in FIG. 235, the actual world estimating unit 102 includes
a conditions setting unit 2331, input image storage unit 2332,
input pixel value acquiring unit 2333, integral component
calculation unit 2334, normal equation generating unit 2335, and
approximation function generating unit 2336.
The conditions setting unit 2331 sets a pixel range (hereafter,
referred to as a tap range) used for estimating the X
cross-sectional waveform F(x) corresponding to a pixel of interest,
and the number of dimensions n of the approximation function
f(x).
The input image storage unit 2332 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel acquiring unit 2333 acquires, of the input images
stored in the input image storage unit 2332, an input image region
corresponding to the tap range set by the conditions setting unit
2231, and supplies this to the normal equation generating unit 2335
as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described. Note that
a specific example of the input pixel value table will be described
later.
Now, the actual world estimating unit 102 calculates the features
w.sub.i of the approximation function f(x) with the least square
method using the above Expression (112) and Expression (113) here,
but the above Expression (112) can be represented such as the
following Expression (115).
.function..times..times..times..function..times..times..times..times..tim-
es..times..times..times..times..times..times..function.
##EQU00076##
In Expression (115), S.sub.i (x.sub.s, x.sub.e) represents the
integral components of the i-dimensional term. That is to say, the
integral components S.sub.i (x.sub.s, x.sub.e) are shown in the
following Expression (116).
.function. ##EQU00077##
The integral component calculation unit 2334 calculates the
integral components S.sub.i (x.sub.s, x.sub.e).
Specifically, the integral components S.sub.i (x.sub.s, x.sub.e)
(however, the value x.sub.s and value x.sub.e are values shown in
the above Expression (112)) shown in Expression (116) may be
calculated as long as the relative pixel positions (x, y), shift
amount C.sub.x (y), and i of the i-dimensional terms are known.
Also, of these, the relative pixel positions (x, y) are determined
by the pixel of interest and the tap range, the shift amount
C.sub.x (y) is determined by the angle .theta. (by the above
Expression (107) and Expression (109)), and the range of i is
determined by the number of dimensions n, respectively.
Accordingly, the integral component calculation unit 2334
calculates the integral components S.sub.i (x.sub.s, x.sub.e) based
on the tap range and the number of dimensions set by the conditions
setting unit 2331, and the angle .theta. of the data continuity
information output from the data continuity detecting unit 101, and
supplies the calculated results to the normal equation generating
unit 2335 as an integral component table.
The normal equation generating unit 2335 generates the above
Expression (112), i.e., a normal equation in the case of obtaining
the features w.sub.i of the right side of Expression (115) with the
least square method using the input pixel value table supplied from
the input pixel value acquiring unit 2333, and the integral
component table supplied from the integral component calculation
unit 2334, and supplies this to the approximation function
generating unit 2336 as a normal equation table. Note that a
specific example of a normal equation will be described later.
The approximation function generating unit 2336 calculates the
respective features w.sub.i of the above Expression (115) (i.e.,
the respective coefficients w.sub.i of the approximation function
f(x) serving as a one-dimensional polynomial) by solving a normal
equation included in the normal equation table supplied from the
normal equation generating unit 2335 using the matrix solution, and
outputs these to the image generating unit 103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40) of the
actual world estimating unit 102 (FIG. 235) which employs the
one-dimensional polynomial approximating method with reference to
the flowchart in FIG. 236.
For example, let us say that an input image, which is a one-frame
input image output from the sensor 2, including the
fine-line-including data region 2302 in FIG. 221 described above
has been already stored in the input image storage unit 2332. Also,
let us say that the data continuity detecting unit 101 has
subjected, at the continuity detection processing in step S101
(FIG. 40), the fine-line-including data region 2302 to the
processing thereof, and has already output the angle .theta. as
data continuity information.
In this case, the conditions setting unit 2331 sets conditions (a
tap range and the number of dimensions) in step S2301 in FIG.
236.
For example, let us say that a tap range 2351 shown in FIG. 237 is
set, and 5 dimensions are set as the number of dimensions.
That is to say, FIG. 237 is a diagram for describing an example of
a tap range. In FIG. 237, the X direction and Y direction are the X
direction and Y direction of the sensor 2 (FIG. 220) respectively.
Also, the tap range 2351 represents a pixel group made up of 20
pixels in total (20 squares in the drawing) of 4 pixels in the X
direction, and also 5 pixels in the Y direction.
Further, as shown in FIG. 237, let us say that a pixel of interest
is set at the second pixel from the left and also the third pixel
from the bottom in the drawing, of the tap range 2351. Also, let us
say that each pixel is denoted with a number l such as shown in
FIG. 237 (l is any integer value of 0 through 19) according to the
relative pixel positions (x, y) from the pixel of interest (a
coordinate value of a pixel-of-interest coordinates system wherein
the center (0, 0) of the pixel of interest is taken as the
origin).
Now, description will return to FIG. 236, wherein in step S2302,
the conditions setting unit 2331 sets a pixel of interest.
In step S2303, the input pixel value acquiring unit 2333 acquires
an input pixel value based on the condition (tap range) set by the
conditions setting unit 2331, and generates an input pixel value
table. That is to say, in this case, the input pixel value
acquiring unit 2333 acquires the fine-line-including data region
2302 (FIG. 225), and generates a table made up of 20 input pixel
values P (l) as an input pixel value table.
Note that in this case, the relation between the input pixel values
P (l) and the above input pixel values P (x, y) is a relation shown
in the following Expression (117). However, in Expression (117),
the left side represents the input pixel values P (l), and the
right side represents the input pixel values P (x, y).
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function..times..times..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..times..t-
imes..function..function..times..times..function..function..times..times..-
function..function..times..times..function..function..times..times..functi-
on..function..times..times..function..function..times..times..function..fu-
nction..times..times..function..function..times..times..function..function-
. ##EQU00078##
In step S2304, the integral component calculation unit 2334
calculates integral components based on the conditions (a tap range
and the number of dimensions) set by the conditions setting unit
2331, and the data continuity information (angle .theta.) supplied
from the data continuity detecting unit 101, and generates an
integral component table.
In this case, as described above, the input pixel values are not P
(x, y) but P (l), and are acquired as the value of a pixel number
l, so the integral component calculation unit 2334 calculates the
above integral components S.sub.i (x.sub.s, x.sub.e) in Expression
(116) as a function of l such as the integral components S.sub.i
(l) shown in the left side of the following Expression (118).
S.sub.i(l)=S.sub.i(x.sub.s,x.sub.e) (118)
Specifically, in this case, the integral components S.sub.i (l)
shown in the following Expression (119) are calculated.
.function..function..times..times..function..function..function..function-
..times..times..function..function..function..function..times..times..func-
tion..function..function..function..times..times..function..function..func-
tion..function..times..times..function..function..function..function..time-
s..times..function..function..function..function..times..times..function..-
function..function..function..times..times..function..function..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..fu-
nction..function..times..times..function..function..function..function..ti-
mes..times..function..function..function..function..times..times..function-
..function..function..function..times..times..function..function..function-
..function..times..times..function..function..function..function..times..t-
imes..function..function..function..function..times..times..function..func-
tion..function..function. ##EQU00079##
Note that in Expression (119), the left side represents the
integral components S.sub.i (l), and the right side represents the
integral components S.sub.i (x.sub.s, x.sub.e). That is to say, in
this case, i is 0 through 5, and accordingly, the 120 S.sub.i (l)
in total of the 20 S.sub.0 (l), 20 S.sub.1 (l), 20 S.sub.2 (l) 20
S.sub.3 (l), 20 S.sub.4 (l), and 20 S.sub.5 (l) are calculated.
More specifically, first the integral component calculation unit
2334 calculates each of the shift amounts C.sub.x (-2), C.sub.x
(-1), C.sub.x (1), and C.sub.x (2) using the angle .theta. supplied
from the data continuity detecting unit 101. Next, the integral
component calculation unit 2334 calculates each of the 20 integral
components S.sub.i (x.sub.s, x.sub.e) shown in the right side of
Expression (118) regarding each of i=0 through 5 using the
calculated shift amounts C.sub.x (-2), C.sub.x (-1), C.sub.x (1),
and C.sub.x (2). That is to say, the 120 integral components
S.sub.i (x.sub.s, x.sub.e) are calculated. Note that with this
calculation of the integral components S.sub.i (x.sub.s, x.sub.e),
the above Expression (116) is used. Subsequently, the integral
component calculation unit 2334 converts each of the calculated 120
integral components S.sub.i (x.sub.s, x.sub.e) into the
corresponding integral components S.sub.i (l) in accordance with
Expression (119), and generates an integral component table
including the converted 120 integral components S.sub.i (l).
Note that the sequence of the processing in step S2303 and the
processing in step S2304 is not restricted to the example in FIG.
236, the processing in step S2304 may be executed first, or the
processing in step S2303 and the processing in step S2304 may be
executed simultaneously.
Next, in step S2305, the normal equation generating unit 2335
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 2333 at the
processing in step S2303, and the integral component table
generated by the integral component calculation unit 2334 at the
processing in step S2304.
Specifically, in this case, the features w.sub.i of the following
Expression (120) corresponding to the above Expression (115) are
calculated using the least square method. A normal equation
corresponding to this is represented as the following Expression
(121).
.function..times..times..function..times..function..times..function..time-
s..times..times..function..times..times..times..function..times..times..ti-
mes..function..times..times..times..function..times..times..times..functio-
n..times..times..times..function..times..times..times..function..times..ti-
mes..times..function..times..times..function..times..function..times..func-
tion..times..function..times..function..times..function.
##EQU00080##
Note that in Expression (121), L represents the maximum value of
the pixel number l in the tap range. n represents the number of
dimensions of the approximation function f(x) serving as a
polynomial. Specifically, in this case, n=5, and L=19.
If we define each matrix of the normal equation shown in Expression
(121) as the following Expressions (122) through (124), the normal
equation is represented as the following Expression (125).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..function..times..function..times..function..times..func-
tion..times..function..times..function..times. ##EQU00081##
As shown in Expression (123), the respective components of the
matrix W.sub.MAT are the features w.sub.i to be obtained.
Accordingly, in Expression (125), if the matrix S.sub.MAT of the
left side and the matrix P.sub.MAT of the right side are
determined, the matrix W.sub.MAT (i.e., features w.sub.i) may by be
calculated with the matrix solution.
Specifically, as shown in Expression (122), the respective
components of the matrix S.sub.MAT may be calculated as long as the
above integral components S.sub.i (l) are known. The integral
components S.sub.i (l) are included in the integral component table
supplied from the integral component calculation unit 2334, so the
normal equation generating unit 2335 can calculate each component
of the matrix S.sub.MAT using the integral component table.
Also, as shown in Expression (124), the respective components of
the matrix P.sub.MAT may be calculated as long as the integral
components S.sub.i (l) and the input pixel values P (l) are known.
The integral components S.sub.i (l) is the same as those included
in the respective components of the matrix S.sub.MAT, also the
input pixel values P (l) are included in the input pixel value
table supplied from the input pixel value acquiring unit 2333, so
the normal equation generating unit 2335 can calculate each
component of the matrix P.sub.MAT using the integral component
table and input pixel value table.
Thus, the normal equation generating unit 2335 calculates each
component of the matrix S.sub.MAT and matrix P.sub.MAT, and outputs
the calculated results (each component of the matrix S.sub.MAT and
matrix P.sub.MAT) to the approximation function generating unit
2336 as a normal equation table.
Upon the normal equation table being output from the normal
equation generating unit 2335, in step S2306, the approximation
function generating unit 2336 calculates the features w.sub.i
(i.e., the coefficients w.sub.i of the approximation function f(x)
serving as a one-dimensional polynomial) serving as the respective
components of the matrix W.sub.MAT in the above Expression (125)
based on the normal equation table.
Specifically, the normal equation in the above Expression (125) can
be transformed as the following Expression (126).
.times. ##EQU00082##
In Expression (126), the respective components of the matrix
W.sub.MAT in the left side are the features w.sub.i to be obtained.
The respective components regarding the matrix S.sub.MAT and matrix
P.sub.MAT are included in the normal equation table supplied from
the normal equation generating unit 2335. Accordingly, the
approximation function generating unit 2336 calculates the matrix
W.sub.MAT by calculating the matrix in the right side of Expression
(126) using the normal equation table, and outputs the calculated
results (features w.sub.i) to the image generating unit 103.
In step S2307, the approximation function generating unit 2336
determines regarding whether or not the processing of all the
pixels has been completed.
In step S2307, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S2303, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S2302 through S2307 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S2307, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
Note that the waveform of the approximation function f(x) generated
with the coefficients (features) w.sub.i thus calculated becomes a
waveform such as the approximation function f3 (x) in FIG. 234
described above.
Thus, with the one-dimensional polynomial approximating method, the
features of the approximation function f(x) serving as a
one-dimensional polynomial are calculated on assumption that a
waveform having the same form as the one-dimensional X
cross-sectional waveform F(x) is continuous in the direction of
continuity. Accordingly, with the one-dimensional polynomial
approximating method, the features of the approximation function
f(x) can be calculated with less amount of calculation processing
than other function approximating methods.
In other words, with the one-dimensional polynomial approximating
method, for example, the multiple detecting elements of the sensor
(for example, detecting elements 2-1 of the sensor 2 in FIG. 220)
each having time-space integration effects project the light
signals in the actual world 1 (for example, an l portion 2301 of
the light signal in the actual world 1 in FIG. 221), and the data
continuity detecting unit 101 in FIG. 219 (FIG. 3) detects
continuity of data (for example, continuity of data represented
with G.sub.f in FIG. 228) in image data (for example, image data
(input image region) 2302 in FIG. 221) made up of multiple pixels
having a pixel value (for example, input pixel values P (x, y)
shown in the respective graphs in FIG. 226) projected by the
detecting elements 2-1, which drop part of continuity (for example,
continuity represented with the gradient G.sub.F in FIG. 228) of
the light signal in the actual world 1.
For example, the actual world estimating unit 102 in FIG. 219 (FIG.
3) estimates the light signal function F by approximating the light
signal function F representing the light signal in the actual world
1 (specifically, X cross-sectional waveform F(x)) with a
predetermined approximation function f(specifically, for example,
the approximation function f.sub.3 (x) in FIG. 234) on condition
that the pixel value (for example, input pixel value P serving as
the left side of the above Expression (112)) of a pixel
corresponding to a position in the one-dimensional direction (for
example, arrow 2311 in FIG. 223, i.e., X direction) of the
time-space directions of image data corresponding to continuity of
data detected by the data continuity detecting unit 101 is the
pixel value (for example, as shown in the right side of Expression
(112), the value obtained by the approximation function f.sub.3 (x)
being integrated in the X direction) acquired by integration
effects in the one-dimensional direction.
Speaking in detail, for example, the actual world estimating unit
102 estimates the light signal function F by approximating the
light signal function F with the approximation function f on
condition that the pixel value of a pixel corresponding to a
distance (for example, shift amounts C.sub.x (y) in FIG. 230) along
in the one-dimensional direction (for example, X direction) from a
line corresponding to continuity of data (for example, a line
(dashed line) corresponding to the gradient G.sub.f in FIG. 230)
detected by the continuity detecting hand unit 101 is the pixel
value (for example, a value obtained by the approximation function
f.sub.3 (x) being integrated in the X direction such as shown in
the right side of Expression (112) with an integral range such as
shown in Expression (112)) acquired by integration effects in the
one-dimensional direction.
Accordingly, with the one-dimensional polynomial approximating
method, the features of the approximation function f(x) can be
calculated with less amount of calculation processing than other
function approximating methods.
Next, description will be made regarding the second function
approximating method with reference to FIG. 238 through FIG.
244.
That is to say, the second function approximating method is a
method wherein the light signal in the actual world 1 having
continuity in the spatial direction represented with the gradient
G.sub.F such as shown in FIG. 238 for example is regarded as a
waveform F(x, y) on the X-Y plane (on the plane level in the X
direction serving as one direction of the spatial directions, and
in the Y direction orthogonal to the X direction), and the waveform
F(x, y) is approximated with the approximation function f(x, y)
serving as a two-dimensional polynomial, thereby estimating the
waveform F(x, y). Accordingly, hereafter, the second function
approximating method is referred to as a two-dimensional polynomial
approximating method.
Note that in FIG. 238, the horizontal direction represents the X
direction serving as one direction of the spatial directions, the
upper right direction represents the Y direction serving as the
other direction of the spatial directions, and the vertical
direction represents the level of light respectively. G.sub.F
represents the gradient as continuity in the spatial direction.
Also, with description of the two-dimensional polynomial
approximating method, let us say that the sensor 2 is a CCD made up
of the multiple detecting elements 2-1 disposed on the plane
thereof, such as shown in FIG. 239.
With the example in FIG. 239, the direction in parallel with a
predetermined side of the detecting elements 2-1 is taken as the X
direction serving as one direction of the spatial directions, and
the direction orthogonal to the X direction is taken as the Y
direction serving as the other direction of the spatial directions.
The direction orthogonal to the X-Y plane is taken as the t
direction serving as the temporal direction.
Also, with the example in FIG. 239, the spatial shape of the
respective detecting elements 2-1 of the sensor 2 is taken as a
square of which one side is 1 in length. The shutter time (exposure
time) of the sensor 2 is taken as 1.
Further, with the example in FIG. 239, the center of one certain
detecting element 2-1 of the sensor 2 is taken as the origin (the
position in the X direction is x=0, and the position in the Y
direction is y=0) in the spatial directions (X direction and Y
direction), and also the intermediate point-in-time of the exposure
time is taken as the origin (the position in the t direction is
t=0) in the temporal direction (t direction).
In this case, the detecting element 2-1 of which the center is in
the origin (x=0, y=0) in the spatial directions subjects the light
signal function F(x, y, t) to integration with a range of -0.5
through 0.5 in the X direction, with a range of -0.5 through 0.5 in
the Y direction, and with a range of -0.5 through 0.5 in the t
direction, and outputs the integral value as the pixel value P.
That is to say, the pixel value P output from the detecting element
2-1 of which the center is in the origin in the spatial directions
is represented with the following Expression (127).
.intg..times..intg..times..intg..times..function..times..times.d.times..t-
imes.d.times..times.d ##EQU00083##
Similarly, the other detecting elements 2-1 output the pixel value
P shown in Expression (127) by taking the center of the detecting
element 2-1 to be processed as the origin in the spatial
directions.
Incidentally, as described above, the two-dimensional polynomial
approximating method is a method wherein the light signal in the
actual world 1 is handled as a waveform F(x, y) such as shown in
FIG. 238 for example, and the two-dimensional waveform F(x, y) is
approximated with the approximation function f(x, y) serving as a
two-dimensional polynomial.
First, description will be made regarding a method representing
such the approximation function f(x, y) with a two-dimensional
polynomial.
As described above, the light signal in the actual world 1 is
represented with the light signal function F(x, y, t) of which
variables are the position on the three-dimensional space x, y, and
z, and point-in-time t. This light signal function F(x, y, t),
i.e., a one-dimensional waveform projected in the X direction at an
arbitrary position y in the Y direction is referred to as an X
cross-sectional waveform F(x), here.
When paying attention to this X cross-sectional waveform F(x), in
the event that the signal in the actual world 1 has continuity in a
certain direction in the spatial directions, it can be conceived
that a waveform having the same form as the X cross-sectional
waveform F(x) continues in the continuity direction. For example,
with the example in FIG. 238, a waveform having the same form as
the X cross-sectional waveform F(x) continues in the direction of
the gradient G.sub.F. In other words, it can be said that the
waveform F(x, y) is formed by a waveform having the same form as
the X cross-sectional waveform F(x) continuing in the direction of
the gradient G.sub.F.
Accordingly, the approximation function f(x, y) can be represented
with a two-dimensional polynomial by considering that the waveform
of the approximation function f(x, y) approximating the waveform
F(x, y) is formed by a waveform having the same form as the
approximation function f(x) approximating the X cross-sectional F
(x) continuing.
Description will be made in more detail regarding the representing
method of the approximation function f(x, y).
For example, let us say that the light signal in the actual world 1
such as shown in FIG. 238 described above, i.e., a light signal
having continuity in the spatial direction represented with the
gradient G.sub.F is detected by the sensor 2 (FIG. 239), and output
as an input image (pixel value).
Further, let us say that as shown in FIG. 240, the data continuity
detecting unit 101 (FIG. 3) subjects an input image region 2401
made up of 20 pixels (in the drawing, 20 squares represented with
dashed line) in total of 4 pixels in the X direction and also 5
pixels in the Y direction, of this input image, to the processing
thereof, and outputs an angle .theta. (angle .theta. generated
between the direction of data continuity represented with the
gradient G.sub.f corresponding to the gradient G.sub.F, and the X
direction) as one of the data continuity information.
Note that with the input image region 2401, the horizontal
direction in the drawing represents the X direction serving as one
direction in the spatial directions, and the vertical direction in
the drawing represents the Y direction serving as the other
direction of the spatial directions.
Also, in FIG. 240, an (x, y) coordinates system is set such that a
pixel in the second pixel from the left, and also the third pixel
from the bottom is taken as a pixel of interest, and the center of
the pixel of interest is taken as the origin (0, 0). A relative
distance (hereafter, referred to as a cross-sectional direction
distance) in the X direction as to the straight line (straight line
having the gradient G.sub.f representing the direction of data
continuity) having an angle .theta. passing through the origin (0,
0) is described as x'.
Further, in FIG. 240, the graph on the right side is a function
wherein an X cross-sectional waveform F(x') is approximated, which
represents an approximation function f(x') serving as an
n-dimensional (n is an arbitrary integer) polynomial. Of the axes
in the graph on the right side, the axis in the horizontal
direction in the drawing represents a cross-sectional direction
distance, and the axis in the vertical direction in the drawing
represents pixel values.
In this case, the approximation function f(x') shown in FIG. 240 is
an n-dimensional polynomial, so is represented as the following
Expression (128).
.function.'.times.'.times.'.times.'.times..times..times..times.'.times..t-
imes. ##EQU00084##
Also, since the angle .theta. is determined, the straight line
having angle .theta. passing through the origin (0, 0) is uniquely
determined, and a position x.sub.1 in the X direction of the
straight line at an arbitrary position y in the Y direction is
represented as the following Expression (129). However, in
Expression (129), s represents cot .theta.. x.sub.1=s.times.y
(129)
That is to say, as shown in FIG. 240, a point on the straight line
corresponding to continuity of data represented with the gradient
G.sub.f is represented with a coordinate value (x.sub.1, y).
The cross-sectional direction distance x' is represented as the
following Expression (130) using Expression (129).
x'=x-x.sub.1=x-s.times.y (130)
Accordingly, the approximation function f(x, y) at an arbitrary
position (x, y) within the input image region 2401 is represented
as the following Expression (131) using Expression (128) and
Expression (130).
.function..times..function..times. ##EQU00085##
Note that in Expression (131), w.sub.i represents coefficients of
the approximation function f(x, y). Note that the coefficients
w.sub.i of the approximation function f including the approximation
function f(x, y) can be evaluated as the features of the
approximation function f. Accordingly, the coefficients w.sub.i of
the approximation function f are also referred to as the features
w.sub.i of the approximation function f.
Thus, the approximation function f(x, y) having a two-dimensional
waveform can be represented as the polynomial of Expression (131)
as long as the angle .theta. is known.
Accordingly, if the actual world estimating unit 102 can calculate
the features w.sub.i of Expression (131), the actual world
estimating unit 102 can estimate the waveform F(x, y) such as shown
in FIG. 238.
Consequently, hereafter, description will be made regarding a
method for calculating the features w.sub.i of Expression
(131).
That is to say, upon the approximation function f(x, y) represented
with Expression (131) being subjected to integration with an
integral range (integral range in the spatial direction)
corresponding to a pixel (the detecting element 2-1 of the sensor 2
(FIG. 239)), the integral value becomes the estimated value
regarding the pixel value of the pixel. It is the following
Expression (132) that this is represented with an equation. Note
that with the two-dimensional polynomial approximating method, the
temporal direction t is regarded as a constant value, so Expression
(132) is taken as an equation of which variables are the positions
x and y in the spatial directions (X direction and Y
direction).
.function..intg..times..intg..times..times..function..times.
##EQU00086##
In Expression (132), P (x, y) represents the pixel value of a pixel
of which the center position is in a position (x, y) (relative
position (x, y) from the pixel of interest) of an input image from
the sensor 2. Also, e represents a margin of error.
Thus, with the two-dimensional polynomial approximating method, the
relation between the input pixel value P (x, y) and the
approximation function f(x, y) serving as a two-dimensional
polynomial can be represented with Expression (132), and
accordingly, the actual world estimating unit 102 can estimate the
two-dimensional function F(x, y) (waveform F(x, y) wherein the
light signal in the actual world 1 having continuity in the spatial
direction represented with the gradient G.sub.F (FIG. 238) is
represented focusing attention on the spatial direction) by
calculating the features w.sub.i with, for example, the least
square method or the like using Expression (132) (by generating the
approximation function f(x, y) by substituting the calculated
features w.sub.i for Expression (130)).
FIG. 241 represents a configuration example of the actual world
estimating unit 102 employing such a two-dimensional polynomial
approximating method.
As shown in FIG. 241, the actual world estimating unit 102 includes
a conditions setting unit 2421, input image storage unit 2422,
input pixel value acquiring unit 2423, integral component
calculation unit 2424, normal equation generating unit 2425, and
approximation function generating unit 2426.
The conditions setting unit 2421 sets a pixel range (tap range)
used for estimating the function F(x, y) corresponding to a pixel
of interest, and the number of dimensions n of the approximation
function f(x, y).
The input image storage unit 2422 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel value acquiring unit 2423 acquires, of the input
images stored in the input image storage unit 2422, an input image
region corresponding to the tap range set by the conditions setting
unit 2421, and supplies this to the normal equation generating unit
2425 as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described. Note that
a specific example of the input pixel value table will be described
later.
Incidentally, as described above, the actual world estimating unit
102 employing the two-dimensional approximating method calculates
the features w.sub.i of the approximation function f(x, y)
represented with the above Expression (131) by solving the above
Expression (132) using the least square method.
Expression (132) can be represented as the following Expression
(137) by using the following Expression (136) obtained by the
following Expressions (133) through (135).
.intg.'.times.d.intg..times..times.d.times..intg..times..times.d.times..f-
unction..intg..times..intg..times..times..times.d.times..times.d.times..in-
tg..times..times..times.d.times..intg..times..times..times..times..times.d-
.times..times..function..times..times..times..function..times..times..time-
s..times..times..times..times..times..times..times..function..times..funct-
ion..times..times..function..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times..times..times..functio-
n. ##EQU00087##
In Expression (137), S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5)
represents the integral components of i-dimensional terms. That is
to say, the integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) are as shown in the following Expression (138).
.function..times..times..times..times..times..times..times..times..functi-
on..times. ##EQU00088##
The integral component calculation unit 2424 calculates the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5).
Specifically, the integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) shown in Expression (138) can be calculated as long as the
relative pixel positions (x, y), the variable s and i of
i-dimensional terms in the above Expression (131) are known. Of
these, the relative pixel positions (x, y) are determined with a
pixel of interest, and a tap range, the variable s is cot .theta.,
which is determined with the angle .theta., and the range of i is
determined with the number of dimensions n respectively.
Accordingly, the integral component calculation unit 2424
calculates the integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) based on the tap range and the number of dimensions set by
the conditions setting unit 2421, and the angle .theta. of the data
continuity information output from the data continuity detecting
unit 101, and supplies the calculated results to the normal
equation generating unit 2425 as an integral component table.
The normal equation generating unit 2425 generates a normal
equation in the case of obtaining the above Expression (132), i.e.,
Expression (137) by the least square method using the input pixel
value table supplied from the input pixel value acquiring unit
2423, and the integral component table supplied from the integral
component calculation unit 2424, and outputs this to the
approximation function generating unit 2426 as a normal equation
table. Note that a specific example of a normal equation will be
described later.
The approximation function generating unit 2426 calculates the
respective features w.sub.i of the above Expression (132) (i.e.,
the coefficients w.sub.i of the approximation function f(x, y)
serving as a two-dimensional polynomial) by solving the normal
equation included in the normal equation table supplied from the
normal equation generating unit 2425 using the matrix solution, and
output these to the image generating unit 103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40) to which
the two-dimensional polynomial approximating method is applied,
with reference to the flowchart in FIG. 242.
For example, let us say that the light signal in the actual world 1
having continuity in the spatial direction represented with the
gradient G.sub.F has been detected by the sensor 2 (FIG. 239), and
has been stored in the input image storage unit 2422 as an input
image corresponding to one frame. Also, let us say that the data
continuity detecting unit 101 has subjected the region 2401 shown
in FIG. 240 described above of the input image to processing in the
continuity detecting processing in step S101 (FIG. 40), and has
output the angle .theta. as data continuity information.
In this case, in step S2401, the conditions setting unit 2421 sets
conditions (a tap range and the number of dimensions).
For example, let us say that a tap range 2441 shown in FIG. 243 has
been set, and also 5 has been set as the number of dimensions.
FIG. 243 is a diagram for describing an example of a tap range. In
FIG. 243, the X direction and Y direction represent the X direction
and Y direction of the sensor 2 (FIG. 239). Also, the tap range
2441 represents a pixel group made up of 20 pixels (20 squares in
the drawing) in total of 4 pixels in the X direction and also 5
pixels in the Y direction.
Further, as shown in FIG. 243, let us say that a pixel of interest
has been set to a pixel, which is the second pixel from the left
and also the third pixel from the bottom in the drawing, of the tap
range 2441. Also, let us say that each pixel is denoted with a
number l such as shown in FIG. 243 (l is any integer value of 0
through 19) according to the relative pixel positions (x, y) from
the pixel of interest (a coordinate value of a pixel-of-interest
coordinates system wherein the center (0, 0) of the pixel of
interest is taken as the origin).
Now, description will return to FIG. 242, wherein in step S2402,
the conditions setting unit 2421 sets a pixel of interest.
In step S2403, the input pixel value acquiring unit 2423 acquires
an input pixel value based on the condition (tap range) set by the
conditions setting unit 2421, and generates an input pixel value
table. That is to say, in this case, the input pixel value
acquiring unit 2423 acquires the input image region 2401 (FIG.
240), generates a table made up of 20 input pixel values P (l) as
an input pixel value table.
Note that in this case, the relation between the input pixel values
P (l) and the above input pixel values P (x, y) is a relation shown
in the following Expression (139). However, in Expression (139),
the left side represents the input pixel values P (l), and the
right side represents the input pixel values P (x, y).
.function..function..times..times..times..times..times..times..function..-
times..times..times..function..times..times..times..function..times..times-
..times..function..times..times..times..function..times..times..times..fun-
ction..times..times..times..function..times..times..times..function..times-
..times..times..function..times..times..times..function..times..times..tim-
es..function..times..times..times..function..times..times..times..function-
..times..times..times..function..times..times..times..function..times..tim-
es..times..function..times..times..times..function..times..times..times..t-
imes..function. ##EQU00089##
In step S2404, the integral component calculation unit 2424
calculates integral components based on the conditions (a tap range
and the number of dimensions) set by the conditions setting unit
2421, and the data continuity information (angle .theta.) supplied
from the data continuity detecting unit 101, and generates an
integral component table.
In this case, as described above, the input pixel values are not P
(x, y) but P (l), and are acquired as the value of a pixel number
l, so the integral component calculation unit 2424 calculates the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) in the
above Expression (138) as a function of l such as the integral
components S.sub.i (l) shown in the left side of the following
Expression (140). S.sub.i(l)=S.sub.i(x-0.5,x+0.5,y-0.5,y+0.5)
(140)
Specifically, in this case, the integral components S.sub.i (l)
shown in the following Expression (141) are calculated.
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function..times..times..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..times..t-
imes..function..function..times..times..function..function..times..times..-
function..function..times..times..function..function..times..times..functi-
on..function..times..times..function..function..times..times..function..fu-
nction..times..times..function..function..times..times..function..function-
. ##EQU00090##
Note that in Expression (141), the left side represents the
integral components S.sub.i (l), and the right side represents the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5). That is
to say, in this case, i is 0 through 5, and accordingly, the 120
S.sub.i (l) in total of the 20 S.sub.0 (l), 20 S.sub.1 (l), 20
S.sub.2 (l), 20 S.sub.3 (l), 20 S.sub.4 (l), and 20 S.sub.5 (l) are
calculated.
More specifically, first the integral component calculation unit
2424 calculates cot .theta. corresponding to the angle .theta.
supplied from the data continuity detecting unit 101, and takes the
calculated result as a variable s. Next, the integral component
calculation unit 2424 calculates each of the 20 integral components
S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) shown in the right side of
Expression (140) regarding each of i=0 through 5 using the
calculated variable s. That is to say, the 120 integral components
S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) are calculated. Note that with
this calculation of the integral components S.sub.i (x-0.5, x+0.5,
y-0.5, y+0.5), the above Expression (138) is used. Subsequently,
the integral component calculation unit 2424 converts each of the
calculated 120 integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) into the corresponding integral components S.sub.i (l) in
accordance with Expression (141), and generates an integral
component table including the converted 120 integral components
S.sub.i (l).
Note that the sequence of the processing in step S2403 and the
processing in step S2404 is not restricted to the example in FIG.
242, the processing in step S2404 may be executed first, or the
processing in step S2403 and the processing in step S2404 may be
executed simultaneously.
Next, in step S2405, the normal equation generating unit 2425
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 2423 at the
processing in step S2403, and the integral component table
generated by the integral component calculation unit 2424 at the
processing in step S2404.
Specifically, in this case, the features w.sub.i are calculated
with the least square method using the above Expression (137)
(however, in Expression (136), the S.sub.i (l) into which the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) are
converted using Expression (140) is used), so a normal equation
corresponding to this is represented as the following Expression
(142).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..times..function..times..function..times..function..time-
s..function..times..function..times..function. ##EQU00091##
Note that in Expression (142), L represents the maximum value of
the pixel number l in the tap range. n represents the number of
dimensions of the approximation function f(x) serving as a
polynomial. Specifically, in this case, n=5, and L=19.
If we define each matrix of the normal equation shown in Expression
(142) as the following Expressions (143) through (145), the normal
equation is represented as the following Expression (146).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..function..times..function..times..function..times..func-
tion..times..function..times..function..times. ##EQU00092##
As shown in Expression (144), the respective components of the
matrix W.sub.MAT are the features w.sub.i to be obtained.
Accordingly, in Expression (146), if the matrix S.sub.MAT of the
left side and the matrix P.sub.MAT of the right side are
determined, the matrix W.sub.MAT may be calculated with the matrix
solution.
Specifically, as shown in Expression (143), the respective
components of the matrix S.sub.MAT may be calculated with the above
integral components S.sub.i (l). That is to say, the integral
components S.sub.i (l) are included in the integral component table
supplied from the integral component calculation unit 2424, so the
normal equation generating unit 2425 can calculate each component
of the matrix S.sub.MAT using the integral component table.
Also, as shown in Expression (145), the respective components of
the matrix P.sub.MAT may be calculated with the integral components
S.sub.i (l) and the input pixel values P (l). That is to say, the
integral components S.sub.i (l) is the same as those included in
the respective components of the matrix S.sub.MAT, also the input
pixel values P (l) are included in the input pixel value table
supplied from the input pixel value acquiring unit 2423, so the
normal equation generating unit 2425 can calculate each component
of the matrix P.sub.MAT using the integral component table and
input pixel value table.
Thus, the normal equation generating unit 2425 calculates each
component of the matrix S.sub.MAT and matrix P.sub.MAT, and outputs
the calculated results (each component of the matrix S.sub.MAT and
matrix P.sub.MAT) to the approximation function generating unit
2426 as a normal equation table.
Upon the normal equation table being output from the normal
equation generating unit 2425, in step S2406, the approximation
function generating unit 2426 calculates the features w.sub.i
(i.e., the coefficients w.sub.i of the approximation function f(x,
y) serving as a two-dimensional polynomial) serving as the
respective components of the matrix W.sub.MAT in the above
Expression (146) based on the normal equation table.
Specifically, the normal equation in the above Expression (146) can
be transformed as the following Expression (147).
.times. ##EQU00093##
In Expression (147), the respective components of the matrix
W.sub.MAT in the left side are the features w.sub.i to be obtained.
The respective components regarding the matrix S.sub.MAT and matrix
P.sub.MAT are included in the normal equation table supplied from
the normal equation generating unit 2425. Accordingly, the
approximation function generating unit 2426 calculates the matrix
W.sub.MAT by calculating the matrix in the right side of Expression
(147) using the normal equation table, and outputs the calculated
results (features w.sub.i) to the image generating unit 103.
In step S2407, the approximation function generating unit 2426
determines regarding whether or not the processing of all the
pixels has been completed.
In step S2407, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S2402, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S2402 through S2407 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S2407, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
As description of the two-dimensional polynomial approximating
method, an example for calculating the coefficients (features)
w.sub.i of the approximation function f(x, y) corresponding to the
spatial directions (X direction and Y direction) has been employed,
but the two-dimensional polynomial approximating method can be
applied to the temporal and spatial directions (X direction and t
direction, or Y direction and t direction) as well.
That is to say, the above example is an example in the case of the
light signal in the actual world 1 having continuity in the spatial
direction represented with the gradient G.sub.F (FIG. 238), and
accordingly, the equation including two-dimensional integration in
the spatial directions (X direction and Y direction), such as shown
in the above Expression (132). However, the concept regarding
two-dimensional integration can be applied not only to the spatial
direction but also to the temporal and spatial directions (X
direction and t direction, or Y direction and t direction).
In other words, with the two-dimensional polynomial approximating
method, even in the case in which the light signal function F(x, y,
t), which needs to be estimated, has not only continuity in the
spatial direction but also continuity in the temporal and spatial
directions (however, X direction and t direction, or Y direction
and t direction), this can be approximated with a two-dimensional
polynomial.
Specifically, for example, in the event that there is an object
moving horizontally in the X direction at uniform velocity, the
direction of movement of the object is represented with like a
gradient V.sub.F in the X-t plane such as shown in FIG. 244. In
other words, it can be said that the gradient V.sub.F represents
the direction of continuity in the temporal and spatial directions
in the X-t plane. Accordingly, the data continuity detecting unit
101 can output movement .theta. such as shown in FIG. 244 (strictly
speaking, though not shown in the drawing, movement .theta. is an
angle generated by the direction of data continuity represented
with the gradient V.sub.f corresponding to the gradient V.sub.F and
the X direction in the spatial direction) as data continuity
information corresponding to the gradient V.sub.F representing
continuity in the temporal and spatial directions in the X-t plane
as well as the above angle .theta. (data continuity information
corresponding to continuity in the spatial directions represented
with the gradient G.sub.F in the X-Y plane).
Accordingly, the actual world estimating unit 102 employing the
two-dimensional polynomial approximating method can calculate the
coefficients (features) w.sub.i of an approximation function f(x,
t) in the same method as the above method by employing the movement
.theta. instead of the angle .theta.. However, in this case, the
equation to be employed is not the above Expression (132) but the
following Expression (148).
.function..intg..times..intg..times..times..function..times..times.d.time-
s.d ##EQU00094##
Note that in Expression (148), s is cot .theta. (however, .theta.
is movement).
Also, an approximation function f(y, t) focusing attention on the
spatial direction Y instead of the spatial direction X can be
handled in the same way as the above approximation function f(x,
t).
Thus, with the two-dimensional polynomial approximating method, for
example, the multiple detecting elements of the sensor (for
example, detecting elements 2-1 of the sensor 2 in FIG. 239) each
having time-space integration effects project the light signals in
the actual world 1 (FIG. 219), and the data continuity detecting
unit 101 in FIG. 219 (FIG. 3) detects continuity of data (for
example, continuity of data represented with G.sub.f in FIG. 240)
in image data (for example, input image in FIG. 219) made up of
multiple pixels having a pixel value projected by the detecting
elements 2-1, which drop part of continuity (for example,
continuity represented with the gradient G.sub.F in FIG. 238) of
the light signal in the actual world 1.
For example, the actual world estimating unit 102 in FIG. 219 (FIG.
3) (FIG. 241 for configuration) estimates the light signal function
F by approximating the light signal function F representing the
light signal in the actual world 1 (specifically, function F(x, y)
in FIG. 238) with an approximation function f(for example,
approximation function f(x, y) shown in Expression (131)) serving
as a polynomial on condition that the pixel value (for example,
input pixel value P (x, y) serving as the left side of the above
Expression (131)) of a pixel corresponding to a position at least
in the two-dimensional direction (for example, spatial direction X
and spatial direction Y in FIG. 238 and FIG. 239) of the time-space
directions of image data corresponding to continuity of data
detected by the data continuity detecting unit 101 is the pixel
value (for example, as shown in the right side of Expression (132),
the value obtained by the approximation function f(x, y) shown in
the above Expression (131) being integrated in the X direction and
Y direction) acquired by integration effects in the two-dimensional
direction.
Speaking in detail, for example, the actual world estimating unit
102 estimates a first function representing the light signals in
the real world by approximating the first function with a second
function serving as a polynomial on condition that the pixel value
of a pixel corresponding to a distance (for example,
cross-sectional direction distance x' in FIG. 240) along in the
two-dimensional direction from a line corresponding to continuity
of data (for example, a line (arrow) corresponding to the gradient
G.sub.f in FIG. 240) detected by the continuity detecting unit 101
is the pixel value acquired by integration effects at least in the
two-dimensional direction.
Thus, the two-dimensional polynomial approximating method takes not
one-dimensional but two-dimensional integration effects into
consideration, so can estimate the light signals in the actual
world 1 more accurately than the one-dimensional polynomial
approximating method.
Next, description will be made regarding the third function
approximating method with reference to FIG. 245 through FIG.
249.
That is to say, the third function approximating method is a method
for estimating the light signal function F(x, y, t) by
approximating the light signal function F(x, y, t) with the
approximation function f(x, y, t) focusing attention on that the
light signal in the actual world 1 having continuity in a
predetermined direction of the temporal and spatial directions is
represented with the light signal function F(x, y, t), for example.
Accordingly, hereafter, the third function approximating method is
referred to as a three-dimensional function approximating
method.
Also, with description of the three-dimensional function
approximating method, let us say that the sensor 2 is a CCD made up
of the multiple detecting elements 2-1 disposed on the plane
thereof, such as shown in FIG. 245.
With the example in FIG. 245, the direction in parallel with a
predetermined side of the detecting elements 2-1 is taken as the X
direction serving as one direction of the spatial directions, and
the direction orthogonal to the X direction is taken as the Y
direction serving as the other direction of the spatial directions.
The direction orthogonal to the X-Y plane is taken as the t
direction serving as the temporal direction.
Also, with the example in FIG. 245, the spatial shape of the
respective detecting elements 2-1 of the sensor 2 is taken as a
square of which one side is 1 in length. The shutter time (exposure
time) of the sensor 2 is taken as 1.
Further, with the example in FIG. 245, the center of one certain
detecting element 2-1 of the sensor 2 is taken as the origin (the
position in the X direction is x=0, and the position in the Y
direction is y=0) in the spatial directions (X direction and Y
direction), and also the intermediate point-in-time of the exposure
time is taken as the origin (the position in the t direction is
t=0) in the temporal direction (t direction).
In this case, the detecting element 2-1 of which the center is in
the origin (x=0, y=0) in the spatial directions subjects the light
signal function F(x, y, t) to integration with a range of -0.5
through 0.5 in the X direction, with a range of -0.5 through 0.5 in
the Y direction, and with a range of -0.5 through 0.5 in the t
direction, and outputs the integral value as the pixel value P.
That is to say, the pixel value P output from the detecting element
2-1 of which the center is in the origin in the spatial directions
is represented with the following Expression (149).
.intg..times..intg..times..intg..times..function..times.d.times.d.times.d
##EQU00095##
Similarly, the other detecting elements 2-1 output the pixel value
P shown in Expression (149) by taking the center of the detecting
element 2-1 to be processed as the origin in the spatial
directions.
Incidentally, as described above, with the three-dimensional
function approximating method, the light signal function F(x, y, t)
is approximated to the three-dimensional approximation function
f(x, y, t).
Specifically, for example, the approximation function f(x, y, t) is
taken as a function having N variables (features), a relational
expression between the input pixel values P (x, y, t) corresponding
to Expression (149) and the approximation function f(x, y, t) is
defined. Thus, in the event that M input pixel values P (x, y, t)
more than N are acquired, N variables (features) can be calculated
from the defined relational expression. That is to say, the actual
world estimating unit 102 can estimate the light signal function
F(x, y, t) by acquiring M input pixel values P (x, y, t), and
calculating N variables (features).
In this case, the actual world estimating unit 102 extracts
(acquires) M input images P (x, y, t), of the entire input image by
using continuity of data included in an input image (input pixel
values) from the sensor 2 as a constraint (i.e., using data
continuity information as to an input image to be output from the
data continuity detecting unit 101). As a result, the prediction
function f(x, y, t) is constrained by continuity of data.
For example, as shown in FIG. 246, in the event that the light
signal function F(x, y, t) corresponding to an input image has
continuity in the spatial direction represented with the gradient
G.sub.F, the data continuity detecting unit 101 results in
outputting the angle .theta. (the angle .theta. generated between
the direction of continuity of data represented with the gradient
G.sub.f (not shown) corresponding to the gradient G.sub.F, and the
X direction) as data continuity information as to the input
image.
In this case, let us say that a one-dimensional waveform wherein
the light signal function F(x, y, t) is projected in the X
direction (such a waveform is referred to as an X cross-sectional
waveform here) has the same form even in the event of projection in
any position in the Y direction.
That is to say, let us say that there is an X cross-sectional
waveform having the same form, which is a two-dimensional (spatial
directional) waveform continuous in the direction of continuity
(angle .theta. direction as to the X direction), and a
three-dimensional waveform wherein such a two-dimensional waveform
continues in the temporal direction t, is approximated with the
approximation function f(x, y, t).
In other words, an X cross-sectional waveform, which is shifted by
a position y in the Y direction from the center of the pixel of
interest, becomes a waveform wherein the X cross-sectional waveform
passing through the center of the pixel of interest is moved
(shifted) by a predetermined amount (amount varies according to the
angle .theta.) in the X direction. Note that hereafter, such an
amount is referred to as a shift amount.
This shift amount can be calculated as follows.
That is to say, the gradient V.sub.f (for example, gradient V.sub.f
representing the direction of data continuity corresponding to the
gradient V.sub.F in FIG. 246) and angle .theta. are represented as
the following Expression (150).
.times..times..theta.dd ##EQU00096##
Note that in Expression (150), dx represents the amount of fine
movement in the X direction, and dy represents the amount of fine
movement in the Y direction as to the dx.
Accordingly, if the shift amount as to the X direction is described
as C.sub.x (y), this is represented as the following Expression
(151).
.function. ##EQU00097##
If the shift amount C.sub.x (y) is thus defined, a relational
expression between the input pixel values P (x, y, t) corresponding
to Expression (149) and the approximation function f(x, y, t) is
represented as the following Expression (152).
.function..intg..times..intg..times..intg..times..function..times.d.times-
.d.times.d ##EQU00098##
In Expression (152), e represents a margin of error. t.sub.s
represents an integration start position in the t direction, and
t.sub.e represents an integration end position in the t direction.
In the same way, y.sub.s represents an integration start position
in the Y direction, and y.sub.e represents an integration end
position in the Y direction. Also, x.sub.s represents an
integration start position in the X direction, and x.sub.e
represents an integration end position in the X direction. However,
the respective specific integral ranges are as shown in the
following Expression (153).
.times..times..times..times..times..times..times..times..function..times.-
.times..function. ##EQU00099##
As shown in Expression (153), it can be represented that an X
cross-sectional waveform having the same form continues in the
direction of continuity (angle .theta. direction as to the X
direction) by shifting an integral range in the X direction as to a
pixel positioned distant from the pixel of interest by (x, y) in
the spatial direction by the shift amount C.sub.x (Y).
Thus, with the three-dimensional function approximating method, the
relation between the pixel values P (x, y, t) and the
three-dimensional approximation function f(x, y, t) can be
represented with Expression (152) (Expression (153) for the
integral range), and accordingly, the light signal function F(x, y,
t) (for example, a light signal having continuity in the spatial
direction represented with the gradient V.sub.F such as shown in
FIG. 246) can be estimated by calculating the N features of the
approximation function f(x, y, t), for example, with the least
square method using Expression (152) and Expression (153).
Note that in the event that a light signal represented with the
light signal function F(x, y, t) has continuity in the spatial
direction represented with the gradient V.sub.F such as shown in
FIG. 246, the light signal function F(x, y, t) may be approximated
as follows.
That is to say, let us say that a one-dimensional waveform wherein
the light signal function F(x, y, t) is projected in the Y
direction (hereafter, such a waveform is referred to as a Y
cross-sectional waveform) has the same form even in the event of
projection in any position in the X direction.
In other words, let us say that there is a two-dimensional (spatial
directional) waveform wherein a Y cross-sectional waveform having
the same form continues in the direction of continuity (angle
.theta. direction as to in the X direction), and a
three-dimensional waveform wherein such a two-dimensional waveform
continues in the temporal direction t is approximated with the
approximation function f(x, y, t).
Accordingly, the Y cross-sectional waveform, which is shifted by x
in the X direction from the center of the pixel of interest,
becomes a waveform wherein the Y cross-sectional waveform passing
through the center of the pixel of interest is moved by a
predetermined shift amount (shift amount changing according to the
angle .theta.) in the Y direction.
This shift amount can be calculated as follows.
That is to say, the gradient G.sub.F is represented as the above
Expression (150), so if the shift amount as to the Y direction is
described as C.sub.y (x), this is represented as the following
Expression (154).
.function..times. ##EQU00100##
If the shift amount C.sub.x (y) is thus defined, a relational
expression between the input pixel values P (x, y, t) corresponding
to Expression (149) and the approximation function f(x, y, t) is
represented as the above Expression (152), as with when the shift
amount C.sub.x (y) is defined.
However, in this case, the respective specific integral ranges are
as shown in the following Expression (155).
.times..times..times..times..function..times..times..function..times..tim-
es..times..times. ##EQU00101##
As shown in Expression (155) (and the above Expression (152)), it
can be represented that a Y cross-sectional waveform having the
same form continues in the direction of continuity (angle .theta.
direction as to the X direction) by shifting an integral range in
the Y direction as to a pixel positioned distant from the pixel of
interest by (x, y), by the shift amount C.sub.x (y).
Thus, with the three-dimensional function approximating method, the
integral range of the right side of the above Expression (152) can
be set to not only Expression (153) but also Expression (155), and
accordingly, the light signal function F(x, y, t) (light signal in
the actual world 1 having continuity in the spatial direction
represented with the gradient G.sub.F) can be estimated by
calculating the n features of the approximation function f(x, y, t)
with, for example, the least square method or the like using
Expression (152) in which Expression (155) is employed as an
integral range.
Thus, Expression (153) and Expression (155), which represent an
integral range, represent essentially the same with only a
difference regarding whether perimeter pixels are shifted in the X
direction (in the case of Expression (153)) or shifted in the Y
direction (in the case of Expression (155)) in response to the
direction of continuity.
However, in response to the direction of continuity (gradient
G.sub.F), there is a difference regarding whether the light signal
function F(x, y, t) is regarded as a group of X cross-sectional
waveforms, or is regarded as a group of Y cross-sectional
waveforms. That is to say, in the event that the direction of
continuity is close to the Y direction, the light signal function
F(x, y, t) is preferably regarded as a group of X cross-sectional
waveforms. On the other hand, in the event that the direction of
continuity is close to the X direction, the light signal function
F(x, y, t) is preferably regarded as a group of Y cross-sectional
waveforms.
Accordingly, it is preferable that the actual world estimating unit
102 prepares both Expression (153) and Expression (155) as an
integral range, and selects any one of Expression (153) and
Expression (155) as the integral range of the right side of the
appropriate Expression (152) in response to the direction of
continuity.
Description has been made regarding the three-dimensional function
method in the case in which the light signal function F(x, y, t)
has continuity (for example, continuity in the spatial direction
represented with the gradient G.sub.F in FIG. 246) in the spatial
directions (X direction and Y direction), but the three-dimensional
function method can be applied to the case in which the light
signal function F(x, y, t) has continuity (continuity represented
with the gradient V.sub.F) in the temporal and spatial directions
(X direction, Y direction, and t direction), as shown in FIG.
247.
That is to say, in FIG. 247, a light signal function corresponding
to a frame #N-1 is taken as F (x, y, #N-1), a light signal function
corresponding to a frame #N is taken as F (x, y, #N), and a light
signal function corresponding to a frame #N+1 is taken as F (x, y,
#N+1).
Note that in FIG. 247, the horizontal direction is taken as the X
direction serving as one direction of the spatial directions, the
upper right diagonal direction is taken as the Y direction serving
as the other direction of the spatial directions, and also the
vertical direction is taken as the t direction serving as the
temporal direction in the drawing.
Also, the frame #N-1 is a frame temporally prior to the frame #N,
the frame #N+1 is a frame temporally following the frame #N. That
is to say, the frame #N-1, frame #N, and frame #N+1 are displayed
in the sequence of the frame #N-1, frame #N, and frame #N+1.
With the example in FIG. 247, a cross-sectional light level along
the direction shown with the gradient V.sub.F (upper right inner
direction from lower left near side in the drawing) is regarded as
generally constant. Accordingly, with the example in FIG. 247, it
can be said that the light signal function F(x, y, t) has
continuity in the temporal and spatial directions represented with
the gradient V.sub.F.
In this case, in the event that a function C (x, y, t) representing
continuity in the temporal and spatial directions is defined, and
also the integral range of the above Expression (152) is defined
with the defined function C (x, y, t), N features of the
approximation function f(x, y, t) can be calculated as with the
above Expression (153) and Expression (155).
The function C (x, y, t) is not restricted to a particular function
as long as this is a function representing the direction of
continuity. However, hereafter, let us say that linear continuity
is employed, and C.sub.x (t) and C.sub.y (t) corresponding to the
shift amount C.sub.x (y) (Expression (151)) and shift amount
C.sub.y (x) (Expression (153)), which are functions representing
continuity in the spatial direction described above, are defined as
a function C (x, y, t) corresponding thereto as follows.
That is to say, if the gradient as continuity of data in the
temporal and spatial directions corresponding to the gradient
G.sub.f representing continuity of data in the above spatial
direction is taken as V.sub.f, and if this gradient V.sub.f is
divided into the gradient in the X direction (hereafter, referred
to as V.sub.fx) and the gradient in the Y direction (hereafter,
referred to as V.sub.fy), the gradient V.sub.fx is represented with
the following Expression (156), and the gradient V.sub.fy is
represented with the following Expression (157), respectively.
dddd ##EQU00102##
In this case, the function C.sub.x (t) is represented as the
following Expression (158) using the gradient V.sub.fx shown in
Expression (156).
.function..times. ##EQU00103##
Similarly, the function C.sub.y (t) is represented as the following
Expression (159) using the gradient V.sub.fy shown in Expression
(157).
.function..times. ##EQU00104##
Thus, upon the function C.sub.x (t) and function C.sub.y (t), which
represent continuity 2511 in the temporal and spatial directions,
being defined, the integral range of Expression (152) is
represented as the following Expression (160).
.times..times..times..times..function..times..times..function..times..tim-
es..function..times..times..function. ##EQU00105##
Thus, with the three-dimensional function approximating method, the
relation between the pixel values P (x, y, t) and the
three-dimensional approximation function f(x, y, t) can be
represented with Expression (152), and accordingly, the light
signal function F(x, y, t) (light signal in the actual world 1
having continuity in a predetermined direction of the temporal and
spatial directions) can be estimated by calculating the n+1
features of the approximation function f(x, y, t) with, for
example, the least square method or the like using Expression (160)
as the integral range of the right side of Expression (152).
FIG. 248 represents a configuration example of the actual world
estimating unit 102 employing such a three-dimensional function
approximating method.
Note that the approximation function f(x, y, t) (in reality, the
features (coefficients) thereof) calculated by the actual world
estimating unit 102 employing the three-dimensional function
approximating method is not restricted to a particular function,
but an n (n=N-1)-dimensional polynomial is employed in the
following description.
As shown in FIG. 248, the actual world estimating unit 102 includes
a conditions setting unit 2521, input image storage unit 2522,
input pixel value acquiring unit 2523, integral component
calculation unit 2524, normal equation generating unit 2525, and
approximation function generating unit 2526.
The conditions setting unit 2521 sets a pixel range (tap range)
used for estimating the light signal function F(x, y, t)
corresponding to a pixel of interest, and the number of dimensions
n of the approximation function f(x, y, t).
The input image storage unit 2522 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel acquiring unit 2523 acquires, of the input images
stored in the input image storage unit 2522, an input image region
corresponding to the tap range set by the conditions setting unit
2521, and supplies this to the normal equation generating unit 2525
as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described.
Incidentally, as described above, the actual world estimating unit
102 employing the three-dimensional function approximating method
calculates the N features (in this case, coefficient of each
dimension) of the approximation function f(x, y) with the least
square method using the above Expression (152) (however, Expression
(153), Expression (156), or Expression (160) for the integral
range).
The right side of Expression (152) can be represented as the
following Expression (161) by calculating the integration
thereof.
.function..times..times..function. ##EQU00106##
In Expression (161), w.sub.i represents the coefficients (features)
of the i-dimensional term, and also S.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e, t.sub.s, t.sub.e) represents the integral
components of the i-dimensional term. However, x.sub.s represents
an integral range start position in the X direction, x.sub.e
represents an integral range end position in the X direction,
y.sub.s represents an integral range start position in the Y
direction, y.sub.e represents an integral range end position in the
Y direction, t.sub.s represents an integral range start position in
the t direction, t.sub.e represents an integral range end position
in the t direction, respectively.
The integral component calculation unit 2524 calculates the
integral components S.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e,
t.sub.s, t.sub.e).
That is to say, the integral component calculation unit 2524
calculates the integral components S.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e, t.sub.s, t.sub.e) based on the tap range and the
number of dimensions set by the conditions setting unit 2521, and
the angle or movement (as the integral range, angle in the case of
using the above Expression (153) or Expression (156), and movement
in the case of using the above Expression (160)) of the data
continuity information output from the data continuity detecting
unit 101, and supplies the calculated results to the normal
equation generating unit 2525 as an integral component table.
The normal equation generating unit 2525 generates a normal
equation in the case of obtaining the above Expression (161) with
the least square method using the input pixel value table supplied
from the input pixel value acquiring unit 2523, and the integral
component table supplied from the integral component calculation
unit 2524, and outputs this to the approximation function
generating unit 2526 as a normal equation table. An example of a
normal equation will be described later.
The approximation function generating unit 2526 calculates the
respective features w.sub.i (in this case, the coefficients w.sub.i
of the approximation function f(x, y) serving as a
three-dimensional polynomial) by solving the normal equation
included in the normal equation table supplied from the normal
equation generating unit 2525 with the matrix solution, and output
these to the image generating unit 103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40) to which
the three-dimensional function approximating method is applied,
with reference to the flowchart in FIG. 249.
First, in step S2501, the conditions setting unit 2521 sets
conditions (a tap range and the number of dimensions).
For example, let us say that a tap range made up of L pixels has
been set. Also, let us say that a predetermined number l (l is any
one of integer values 0 through L-1) is appended to each of the
pixels.
Next, in step S2502, the conditions setting unit 2521 sets a pixel
of interest.
In step S2503, the input pixel value acquiring unit 2523 acquires
an input pixel value based on the condition (tap range) set by the
conditions setting unit 2521, and generates an input pixel value
table. In this case, a table made up of L input pixel values P (x,
y, t) is generated. Here, let us say that each of the L input pixel
values P (x, y, t) is described as P (l) serving as a function of
the number l of the pixel thereof. That is to say, the input pixel
value table becomes a table including L P (l).
In step S2504, the integral component calculation unit 2524
calculates integral components based on the conditions (a tap range
and the number of dimensions) set by the conditions setting unit
2521, and the data continuity information (angle or movement)
supplied from the data continuity detecting unit 101, and generates
an integral component table.
However, in this case, as described above, the input pixel values
are not P (x, y, t) but P (l), and are acquired as the value of a
pixel number l, so the integral component calculation unit 2524
results in calculating the integral components S.sub.i (x.sub.s,
x.sub.e, y.sub.s, y.sub.e, t.sub.s, t.sub.e) in the above
Expression (161) as a function of l such as the integral components
S.sub.i (l). That is to say, the integral component table becomes a
table including L.times.i S.sub.i (l).
Note that the sequence of the processing in step S2503 and the
processing in step S2504 is not restricted to the example in FIG.
249, so the processing in step S2504 may be executed first, or the
processing in step S2503 and the processing in step S2504 may be
executed simultaneously.
Next, in step S2505, the normal equation generating unit 2525
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 2523 at the
processing in step S2503, and the integral component table
generated by the integral component calculation unit 2524 at the
processing in step S2504. Specifically, in this case, the features
w.sub.i of the following Expression (162) corresponding to the
above Expression (161) are calculated using the least square
method. A normal equation corresponding to this is represented as
the following Expression (163).
.function..times..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..function..times..function..times..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction. ##EQU00107##
If we define each matrix of the normal equation shown in Expression
(163) as the following Expressions (164) through (166), the normal
equation is represented as the following Expression (167).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..function..times..function..times..function..times..func-
tion..times..function..times..function..times. ##EQU00108##
As shown in Expression (165), the respective components of the
matrix W.sub.MAT are the features w.sub.i to be obtained.
Accordingly, in Expression (167), if the matrix S.sub.MAT of the
left side and the matrix P.sub.MAT of the right side are
determined, the matrix W.sub.MAT (i.e., features w.sub.i) may by be
calculated with the matrix solution.
Specifically, as shown in Expression (164), the respective
components of the matrix S.sub.MAT may be calculated as long as the
above integral components S.sub.i (l) are known. The integral
components S.sub.i (l) are included in the integral component table
supplied from the integral component calculation unit 2524, so the
normal equation generating unit 2525 can calculate each component
of the matrix S.sub.MAT using the integral component table.
Also, as shown in Expression (166), the respective components of
the matrix P.sub.MAT may be calculated as long as the integral
components S.sub.i (l) and the input pixel values P (l) are known.
The integral components S.sub.i (l) is the same as those included
in the respective components of the matrix S.sub.MAT, also the
input pixel values P (l) are included in the input pixel value
table supplied from the input pixel value acquiring unit 2523, so
the normal equation generating unit 2525 can calculate each
component of the matrix P.sub.MAT using the integral component
table and input pixel value table.
Thus, the normal equation generating unit 2525 calculates each
component of the matrix S.sub.MAT and matrix P.sub.MAT, and outputs
the calculated results (each component of the matrix S.sub.MAT and
matrix P.sub.MAT) to the approximation function generating unit
2526 as a normal equation table.
Upon the normal equation table being output from the normal
equation generating unit 2526, in step S2506, the approximation
function generating unit 2526 calculates the features w.sub.i
(i.e., the coefficients w.sub.i of the approximation function f(x,
y, t)) serving as the respective components of the matrix W.sub.MAT
in the above Expression (167) based on the normal equation
table.
Specifically, the normal equation in the above Expression (167) can
be transformed as the following Expression (168).
.times. ##EQU00109##
In Expression (168), the respective components of the matrix
W.sub.MAT in the left side are the features w.sub.i to be obtained.
The respective components regarding the matrix S.sub.MAT and matrix
P.sub.MAT are included in the normal equation table supplied from
the normal equation generating unit 2525. Accordingly, the
approximation function generating unit 2526 calculates the matrix
W.sub.MAT by calculating the matrix in the right side of Expression
(168) using the normal equation table, and outputs the calculated
results (features w.sub.i) to the image generating unit 103.
In step S2507, the approximation function generating unit 2526
determines regarding whether or not the processing of all the
pixels has been completed.
In step S2507, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S2502, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S2502 through S2507 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S5407, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
As described above, the three-dimensional function approximating
method takes three-dimensional integration effects in the temporal
and spatial directions into consideration instead of
one-dimensional or two-dimensional integration effects, and
accordingly, can estimate the light signals in the actual world 1
more accurately than the one-dimensional polynomial approximating
method and two-dimensional polynomial approximating method.
In other words, with the three-dimensional function approximating
method, for example, the actual world estimating unit 102 in FIG.
219 (FIG. 3) (for example, FIG. 248 for configuration) estimates
the light signal function F by approximating the light signal
function F representing the light signal in the actual world
(specifically, for example, the light signal function F(x, y, t) in
FIG. 246 and FIG. 247) with a predetermined approximation function
f(specifically, for example, the approximation function f(x, y, t)
in the right side of Expression (152)), on condition that the
multiple detecting elements of the sensor (for example, detecting
elements 2-1 of the sensor 2 in FIG. 245) each having time-space
integration effects project the light signals in the actual world
1, of the input image made up of multiple pixels having a pixel
value projected by the detecting elements, which drop part of
continuity (for example, continuity represented with the gradient
G.sub.F in FIG. 246, or represented with the gradient V.sub.F in
FIG. 247) of the light signal in the actual world 1, the above
pixel value (for example, input pixel values P (x, y, z) in the
left side of Expression (153)) of the above pixel corresponding to
at least a position in the one-dimensional direction (for example,
three-dimensional directions of the spatial direction X, spatial
direction Y, and temporal direction t in FIG. 247) of the
time-space directions is a pixel value (for example, a value
obtained by the approximation function f(x, y, t) being integrated
in three dimensions of the X direction, Y direction, and t
direction, such as shown in the right side of the above Expression
(153)) acquired by at least integration effects in the
one-dimensional direction.
Further, for example, in the event that the data continuity
detecting unit 101 in FIG. 219 (FIG. 3) detects continuity of input
image data, the actual world estimating unit 102 estimates the
light signal function F by approximating the light signal function
F with the approximation function f on condition that the pixel
value of a pixel corresponding to at least a position in the
one-dimensional direction of the time-space directions of the image
data corresponding to continuity of data detected by the data
continuity detecting unit 101 is the pixel value acquired by at
least integration effects in the one-dimensional direction.
Speaking in detail, for example, the actual world estimating unit
102 estimates the light signal function by approximating the light
signal function F with the approximation function f on condition
that the pixel value of a pixel corresponding to a distance (for
example, shift amounts C.sub.x (y) in the above Expression (151))
along at least in the one-dimensional direction from a line
corresponding to continuity of data detected by the continuity
detecting unit 101 is the pixel value (for example, a value
obtained by the approximation function f(x, y, t) being integrated
in three dimensions of the X direction, Y direction, and t
direction, such as shown in the right side of Expression (152) with
an integral range such as shown in the above Expression (153))
acquired by at least integration effects in the one-dimensional
direction.
Accordingly, the three-dimensional function approximating method
can estimate the light signals in the actual world 1 more
accurately.
Next, description will be made regarding another example of an
extracting method for extracting the data 162 in the case of the
actual world estimating unit 102 approximating the actual world 1
signals having continuity with the model 161, with reference to
FIG. 250 through FIG. 259.
With the following example, the pixel value of each pixel to which
weight according to the level of importance of each pixel is added
is extracted, the extracted value is used as the data 162 (FIG. 7),
and the actual world 1 signals are approximated with the model 161
(FIG. 7).
Specifically, for example, let us say that an input image 2701 such
as shown in FIG. 250 is input to the actual world estimating unit
102 (FIG. 3) as an input image from the sensor 2 (FIG. 1).
In FIG. 250, the horizontal axis in the drawing represents the
X-direction which is one spatial direction, and the vertical
direction in the drawing represents the Y-direction which is
another spatial direction.
Also, the input image 2701 is made up of pixel values (expressed
with the hatched line in the drawing, but actually, data having one
value) of 7.times.16 pixels (square in the drawing) each having
pixel widths (vertical width and horizontal width) L.sub.c.
A pixel of interest is taken as a pixel having a pixel value 2701-1
(hereafter, the pixel having the pixel value 2701-1 is referred to
as a pixel of interest 2701-1), and the direction of data
continuity in the pixel of interest 2701-1 is expressed with a
gradient G.sub.f.
FIG. 251 illustrates the difference between the level of the actual
world 1 light signals at the center of the pixel of interest 2701-1
and the level of the actual world 1 light signals in a
cross-sectional direction distance x' (hereafter, referred to as
difference of levels). That is to say, the axis in the horizontal
direction in the drawing represents the cross-sectional direction
distance x', and the axis in the vertical direction in the drawing
represents difference of levels. Note that the numeric values in
the axis in the horizontal direction are appended with the pixel
widths L.sub.c as 1 in length.
Now, description will be made regarding the cross-sectional
direction distance x' with reference to FIG. 252 and FIG. 253.
FIG. 252 illustrates a 5.times.5 pixel block centered on the pixel
of interest 2701-1, of the input image 2701 shown in FIG. 250. In
FIG. 252 as well, as with FIG. 250, the horizontal axis in the
drawing represents the X-direction which is one spatial direction,
and the vertical direction in the drawing represents the
Y-direction which is another spatial direction.
At this time, for example, if we say that the center of the pixel
of interest 2701-1 is taken as the origin (0, 0) in the spatial
directions, and a straight line, which passes through the origin,
in parallel with the direction of data continuity (with the example
shown in FIG. 252, the direction of data continuity represented
with the gradient G.sub.f) is drawn, the relative distance in the X
direction as to the straight line is referred to as the
cross-sectional direction distance x'. With the example shown in
FIG. 252, the cross-sectional direction distance x' in the center
point of the pixel 2701-2 two pixels apart from the pixel of
interest 2701-1 in the Y direction is illustrated.
FIG. 253 is a figure representing the cross-sectional direction
distance of each pixel within the block shown in FIG. 252 of the
input image 2701 shown in FIG. 250. That is to say, in FIG. 253,
the value marked within each pixel in the input image 2701 (square
region of 5.times.5=25 pixels in the drawing) represents the
cross-sectional direction distance at the corresponding pixel. For
example, the cross-sectional direction distance X.sub.n' at the
pixel 2701-2 is -2.beta..
Note that, as described above, each pixel widths L.sub.c are
defined with the pixel width of 1 in both the X-direction and the
Y-direction. Furthermore, the X-direction is defined with the
positive direction matching the right direction in the drawing.
Also, .beta. represents the cross-sectional direction distance at
the pixel 2701-3 adjacent to the pixel of interest 2701-1 in the
Y-direction (adjacent thereto downward in the drawing). In the
event that the data continuity detecting unit 101 supplies the
angle .theta. (the angle .theta. between the direction of data
continuity represented with the gradient G.sub.f and the
X-direction) as shown in FIG. 253 as the data continuity
information, and accordingly, this value .beta. can be obtained
with ease using the following Expression (169).
.beta..times..times..theta. ##EQU00110##
Now, description will return to FIG. 251. It is difficult to draw
actual difference of levels, so with the example shown in FIG. 251,
a higher-resolution image (not shown) than the input image 2701,
corresponding to the input image 2701 shown in FIG. 250, is created
beforehand. Of the pixels in the high-resolution image, the
difference between the pixel value of the pixel (the pixel of the
high-resolution image) positioned on the general center of the
pixel of interest 2701-1 of the input image 2701 and each pixel
(the pixel of the high-resolution image) positioned on the straight
line, which is a straight line in parallel with the spatial
direction X, passing through the center of the pixel of interest
2701-1 of the input image 2701 is plotted as difference of
levels.
In FIG. 251, as shown with the difference of levels plotted, the
region (hereafter, such a region is referred to as a continuity
region in the description of weighting) having data continuity
represented with the gradient G.sub.f exists in a range between
around -0.5 and around 1.5 of the cross-sectional direction
distance x'.
Accordingly, the smaller cross-sectional direction distance x' a
pixel (the pixel of the input image 2701) has, the higher
probability of including the continuity region is. That is to say,
we can say that the pixel value of the pixel (the pixel of the
input image 2701) of which the cross-sectional direction distance
x' is small is high in the level of importance as the data 162 in
the event that the actual world estimating unit 102 approximates
the actual world 1 signals having continuity with the model
161.
Conversely, the greater cross-sectional direction distance x' a
pixel (the pixel of the input image 2701) has, the lower
probability of including the continuity region is. That is to say,
we can say that the pixel value of the pixel (the pixel of the
input image 2701) of which the cross-sectional direction distance
x' is great is low in the level of importance as the data 162 in
the event that the actual world estimating unit 102 approximates
the actual world 1 signals having continuity with the model
161.
The above relationship of level of importance can be adapted to all
of the input images from the sensor 2 (FIG. 1) as well as the input
image 2701.
To this end, in the event of approximating the actual world 1
signals having continuity with the model 161, the actual world
estimating unit 102 subjects the pixel value of each pixel (the
pixel of the input image from the sensor 2) to weighting according
to the cross-sectional direction distance x' thereof to extract the
weighted pixel value, and the extracted value (weighted pixel
value) can be employed as the data 162. That is to say, in the
event that the pixel value of the input image is extracted as the
data 162, the pixel value is extracted such that the greater the
cross-sectional direction distance x' thereof is, the smaller the
weight thereof is, as shown in FIG. 251.
Further, as shown in FIG. 254, in the event of approximating the
actual world 1 signals having continuity with the model 161, the
actual world estimating unit 102 subjects the pixel value of each
pixel (the pixel of the input image from the sensor 2, the pixel of
the input image 2701 in the example shown in FIG. 254) to weighting
according to the spatial correlation thereof (i.e., according to
the distance of the direction of continuity represented with the
gradient G.sub.f from the pixel of interest 2701-1) to extract the
weighted pixel value, and the extracted value (weighted pixel
value) can be employed as the data 162. That is to say, in the
event that the pixel value of the input image is extracted as the
data 162, the pixel value is extracted such that the smaller the
spatial correlation thereof is (the greater the distance of the
direction of continuity represented with the gradient G.sub.f), the
smaller the weight thereof is, as shown in FIG. 254. Note that FIG.
254 illustrates the same input image 2701 as that shown in FIG.
250.
With the above two types of weighting (weighting shown in FIG. 251
and weighting shown in FIG. 254), either one may be employed, or
both may be employed simultaneously. Note that in the event that
both are employed simultaneously, a finally employed weight
calculating method is not restricted to any particular one. For
example, as final weight, the product of both weight may be
employed, or the weight corrected according to the distance of
direction of data continuity represented with the gradient G.sub.f
as to the weight determined by the weighting shown in FIG. 251 may
be employed (e.g., each time the distance of direction of data
continuity increases by 1, the weight decreases by a predetermined
value).
The actual world estimating unit 102 extracts the pixel value of
each pixel using the weight thus determined, and employs the
weighted pixel value as the data 162, thereby enabling the model
161 closer to the actual world 1 signal to be generated.
Specifically, for example, the actual world estimating unit 102 can
estimate the actual world 1 signals by computing the features of an
approximation function serving as the model 161 (i.e., each
component of the matrix W.sub.MAT) using an normal equation
represented with S.sub.MATW.sub.MAT=P.sub.MAT (i.e., the least
square method) as well, as described above.
In this case, of the input image, if we say that the weight
corresponding to each pixel having a pixel number l (l is any
integer number of 1 through M) is written as v.sub.1, the actual
world estimating unit 102 can use the matrix shown in the following
Expression (170) as the matrix S.sub.MAT, and also use the matrix
shown in the following Expression (171) as the matrix
P.sub.MAT.
.times..times..function..times..function..times..times..function..times..-
function..times..times..function..times..function..times..times..function.-
.times..function..times..times..function..times..function..times..times..f-
unction..times..function..times..times..function..times..function..times..-
times..function..times..function..times..times..function..times..function.-
.times..times..function..times..function..times..times..function..times..f-
unction..times..times..function..times..function. ##EQU00111##
Thus, the actual world estimating unit 102, which employs the least
square method such as the above function approximation technique
(FIG. 219), can compute the features of the approximation function
closer to the actual world 1 signals by using the matrices
including weight (i.e., the above Expression (170) and above
Expression (171)) as compared to the case of using the matrix shown
in the above Expression (13) as the matrix S.sub.MAT, and also
using the matrix shown in the above Expression (15) as the matrix
P.sub.MAT.
That is to say, the actual world estimating unit 102, which employs
the least square method, can compute the features of the
approximation function closer to the actual world 1 signals by
further executing the above weighting processing (as a matrix used
in a normal equation, such as shown in Expression (170) and
Expression (171), simply by using the matrices including the weight
v.sub.1) without changing the configuration thereof.
Specifically, for example, FIG. 255 illustrates an example of an
image generated by the actual world estimating unit 102 generating
an approximation function using the matrices not including the
weight v.sub.1 (e.g., the above Expression (13) and Expression
(15)) as matrices in a normal equation (computing the features of
an approximation function), and the image generating unit 103 (FIG.
3) reintegrating the approximation function.
On the other hand, FIG. 256 illustrates an example of an image
(image corresponding to FIG. 255) generated by the actual world
estimating unit 102 generating an approximation function using the
matrices including the weight v.sub.1 (e.g., the above Expression
(170) and Expression (171)) as matrices in a normal equation
(computing the features of an approximation function), and the
image generating unit 103 reintegrating the approximation
function.
When comparing the image shown in FIG. 255 with the image shown in
FIG. 256, for example, both an image region 2711 shown in FIG. 255
and an image region 2712 shown in FIG. 256 express a part of the
tip of a fork (the same portion).
In the image region 2711 shown in FIG. 255, discontinuous multiple
lines are displayed so as to be overlaid, but in the image region
2712 shown in FIG. 256, approximately one continuous line is
displayed.
When considering that the tip of the fork is actually formed
continuously (one continuous line as viewed from the eyes of a
human), we can say that the image region 2712 shown in FIG. 256
reproduces the actual world 1 signals, i.e., the image of the tip
of the fork truer than the image region 2711 shown in FIG. 255.
Also, FIG. 257 illustrates another example of an image (image
different from FIG. 255) generated by the actual world estimating
unit 102 generating an approximation function using the matrices
not including the weight v.sub.1 (e.g., the above Expression (13)
and Expression (15)) as matrices in a normal equation (computing
the features of an approximation function), and the image
generating unit 103 reintegrating the approximation function.
Conversely, FIG. 258 illustrates another example of an image (image
corresponding to FIG. 257, but an example different from the image
shown in FIG. 256) generated by the actual world estimating unit
102 generating an approximation function using the matrices
including the weight v.sub.1 (e.g., the above Expression (170) and
Expression (171)) as matrices in a normal equation (computing the
features of an approximation function), and the image generating
unit 103 reintegrating the approximation function.
When comparing the image shown in FIG. 257 with the image shown in
FIG. 258, for example, both an image region 2713 shown in FIG. 257
and an image region 2714 shown in FIG. 258 express a part of a beam
(the same portion).
In the image region 2713 shown in FIG. 257, discontinuous multiple
lines are displayed so as to be overlaid, but in the image region
2714 shown in FIG. 258, approximately one continuous line is
displayed.
When considering that the beam is actually formed continuously (one
continuous line as viewed from the eyes of a human), we can say
that the image region 2714 shown in FIG. 258 reproduces the actual
world 1 signals, i.e., the image of the beam, truer than the image
region 2713 shown in FIG. 257.
According to the above arrangement, continuity of data in image
data made up of multiple pixels having a pixel value on which the
real world light signals are projected by the multiple detecting
elements of the sensor each having spatio-temporal integration
effects, of which a part of continuity of the real world light
signals has been dropped, is detected, weight is added to each
pixel within the image data according to at least a distance in the
one-dimensional direction of the time-space directions from a pixel
of interest within the image data, corresponding to the detected
continuity of data, assuming that the pixel values weighted of the
pixels corresponding to positions in at least one dimensional
direction are pixel values acquired by the integration effects in
at least one dimensional direction, a first function representing
the real world light signals is approximated with a second function
serving as a polynomial, thereby estimating the first function, and
accordingly, the image can be expressed in a truer manner.
Next, description will be made regarding an embodiment of the image
generating unit 103 (FIG. 3) with reference to FIG. 259 through
FIG. 280.
FIG. 259 is a diagram for describing the principle of the present
embodiment.
As shown in FIG. 259, the present embodiment is based on condition
that the actual world estimating unit 102 employs a function
approximating method. That is to say, let us say that the signals
in the actual world 1 (distribution of light intensity) serving as
an image cast in the sensor 2 are represented with a predetermined
function F, it is an assumption for the actual world estimating
unit 102 to estimate the function F by approximating the function F
with a predetermined function f using the input image (pixel value
P) output from the sensor 2 and the data continuity information
output from the data continuity detecting unit 101.
Note that hereafter, with description of the present embodiment,
the signals in the actual world 1 serving as an image are
particularly referred to as light signals, and the function F is
particularly referred to as a light signal function F. Also, the
function f is particularly referred to as an approximation function
f.
With the present embodiment, the image generating unit 103
integrates the approximation function f with a predetermined
time-space region using the data continuity information output from
the data continuity detecting unit 101, and the actual world
estimating information (in the example in FIG. 259, the features of
the approximation function f) output from the actual world
estimating unit 102 based on such an assumption, and outputs the
integral value as an output pixel value M (output image). Note that
with the present embodiment, an input pixel value is described as
P, and an output pixel value is described as M in order to
distinguish an input image pixel from an output image pixel.
In other words, upon the light signal function F being integrated
once, the light signal function F becomes an input pixel value P,
the light signal function F is estimated from the input pixel value
P (approximated with the approximation function f), the estimated
light signal function F (i.e., approximation function f) is
integrated again, and an output pixel value M is generated.
Accordingly, hereafter, integration of the approximation function f
executed by the image generating unit 103 is referred to as
reintegration. Also, the present embodiment is referred to as a
reintegration method.
Note that as described later, with the reintegration method, the
integral range of the approximation function f in the event that
the output pixel value M is generated is not restricted to the
integral range of the light signal function F in the event that the
input pixel value P is generated (i.e., the vertical width and
horizontal width of the detecting element of the sensor 2 for the
spatial direction, the exposure time of the sensor 2 for the
temporal direction), an arbitrary integral range may be
employed.
For example, in the event that the output pixel value M is
generated, varying the integral range in the spatial direction of
the integral range of the approximation function f enables the
pixel pitch of an output image according to the integral range
thereof to be varied. That is to say, creation of spatial
resolution is available.
In the same way, for example, in the event that the output pixel
value M is generated, varying the integral range in the temporal
direction of the integral range of the approximation function f
enables creation of temporal resolution.
Hereafter, description will be made individually regarding three
specific methods of such a reintegration method with reference to
the drawings.
That is to say, three specific methods are reintegration methods
corresponding to three specific methods of the function
approximating method (the above three specific examples of the
embodiment of the actual world estimating unit 102)
respectively.
Specifically, the first method is a reintegration method
corresponding to the above one-dimensional polynomial approximating
method (one method of the function approximating method).
Accordingly, with the first method, one-dimensional reintegration
is performed, so hereafter, such a reintegration method is referred
to as a one-dimensional reintegration method.
The second method is a reintegration method corresponding to the
above two-dimensional polynomial approximating method (one method
of the function approximating method). Accordingly, with the second
method, two-dimensional reintegration is performed, so hereafter,
such a reintegration method is referred to as a two-dimensional
reintegration method.
The third method is a reintegration method corresponding to the
above three-dimensional function approximating method (one method
of the function approximating method). Accordingly, with the third
method, three-dimensional reintegration is performed, so hereafter,
such a reintegration method is referred to as a three-dimensional
reintegration method.
Hereafter, description will be made regarding each details of the
one-dimensional reintegration method, two-dimensional reintegration
method, and three-dimensional reintegration method in this
order.
First, the one-dimensional reintegration method will be
described.
With the one-dimensional reintegration method, it is an assumption
that the approximation function f(x) is generated using the
one-dimensional polynomial approximating method.
That is to say, it is an assumption that a one-dimensional waveform
(with description of the reintegration method, a waveform projected
in the X direction of such a waveform is referred to as an X
cross-sectional waveform F(x)) wherein the light signal function
F(x, y, t) of which variables are positions x, y, and z on the
three-dimensional space, and a point-in-time t is projected in a
predetermined direction (for example, X direction) of the X
direction, Y direction, and z direction serving as the spatial
direction, and t direction serving as the temporal direction, is
approximated with the approximation function f(x) serving as an
n-dimensional (n is an arbitrary integer) polynomial.
In this case, with the one-dimensional reintegration method, the
output pixel value M is calculated such as the following Expression
(172).
.times..intg..times..function..times.d ##EQU00112##
Note that in Expression (172), x.sub.s represents an integration
start position, x.sub.e represents an integration end position.
Also, G.sub.e represents a predetermined gain.
Specifically, for example, let us say that the actual world
estimating unit 102 has already generated the approximation
function f(x) (the approximation function f(x) of the X
cross-sectional waveform F(x)) such as shown in FIG. 260 with a
pixel 3101 (pixel 3101 corresponding to a predetermined detecting
element of the sensor 2) such as shown in FIG. 260 as a pixel of
interest.
Note that with the example in FIG. 260, the pixel value (input
pixel value) of the pixel 3101 is taken as P, and the shape of the
pixel 3101 is taken as a square of which one side is 1 in length.
Also, of the spatial directions, the direction in parallel with one
side of the pixel 3101 (horizontal direction in the drawing) is
taken as the X direction, and the direction orthogonal to the X
direction (vertical direction in the drawing) is taken as the Y
direction.
Also, on the lower side in FIG. 260, the coordinates system
(hereafter, referred to as a pixel-of-interest coordinates system)
in the spatial directions (X direction and Y direction) of which
the origin is taken as the center of the pixel 3101, and the pixel
3101 in the coordinates system are shown.
Further, on the upward direction in FIG. 260, a graph representing
the approximation function f(x) at y=0 (y is a coordinate value in
the Y direction in the pixel-of-interest coordinates system shown
on the lower side in the drawing) is shown. In this graph, the axis
in parallel with the horizontal direction in the drawing is the
same axis as the x axis in the X direction in the pixel-of-interest
coordinates system shown on the lower side in the drawing (the
origin is also the same), and also the axis in parallel with the
vertical direction in the drawing is taken as an axis representing
pixel values.
In this case, the relation of the following Expression (173) holds
between the approximation function f(x) and the pixel value P of
the pixel 3101.
.intg..times..function..times.d ##EQU00113##
Also, as shown in FIG. 260, let us say that the pixel 3101 has
continuity of data in the spatial direction represented with the
gradient G.sub.f. Further, let us say that the data continuity
detecting unit 101 (FIG. 259) has already output the angle .theta.
such as shown in FIG. 260 as data continuity information
corresponding to continuity of data represented with the gradient
G.sub.f.
In this case, for example, with the one-dimensional reintegration
method, as shown in FIG. 261, four pixels 3111 through 3114 can be
newly created in a range of -0.5 through 0.5 in the X direction,
and also in a range of -0.5 through 0.5 in the Y direction (in the
range where the pixel 3101 in FIG. 260 is positioned).
Note that on the lower side in FIG. 261, the same pixel-of-interest
coordinates system as that in FIG. 260, and the pixels 3111 through
3114 in the pixel-of-interest coordinates system thereof are shown.
Also, on the upper side in FIG. 261, the same graph (graph
representing the approximation function f(x) at y=0) as that in
FIG. 260 is shown.
Specifically, as shown in FIG. 261, with the one-dimensional
reintegration method, calculation of the pixel value M (1) of the
pixel 3111 using the following Expression (174), calculation of the
pixel value M (2) of the pixel 3112 using the following Expression
(175), calculation of the pixel value M (3) of the pixel 3113 using
the following Expression (176), and calculation of the pixel value
M (4) of the pixel 3114 using the following Expression (177) are
available respectively.
.function..times..intg..times..times..times..times..times..function..time-
s.d.function..times..intg..times..times..times..times..times..function..ti-
mes.d.function..times..intg..times..times..times..times..times..function..-
times.d.function..times..intg..times..times..times..times..times..function-
..times.d ##EQU00114##
Note that x.sub.s1 in Expression (174), x.sub.s2 in Expression
(175), x.sub.s3 in Expression (176), and x.sub.s4 in Expression
(177) each represent the integration start position of the
corresponding expression. Also, x.sub.e1 in Expression (174),
x.sub.e2 in Expression (175), x.sub.e3 in Expression (176), and
x.sub.e4 in Expression (177) each represent the integration end
position of the corresponding expression.
The integral range in the right side of each of Expression (174)
through Expression (177) becomes the pixel width (length in the X
direction) of each of the pixel 3111 through pixel 3114. That is to
say, each of x.sub.e1-x.sub.s1, x.sub.e2-x.sub.s2,
x.sub.e3-x.sub.s3, and x.sub.e4-x.sub.s4 becomes 0.5.
However, in this case, it can be conceived that a one-dimensional
waveform having the same form as that in the approximation function
f(x) at y=0 continues not in the Y direction but in the direction
of data continuity represented with the gradient G.sub.f (i.e.,
angle .theta. direction) (in fact, a waveform having the same form
as the X cross-sectional waveform F(x) at y=0 continues in the
direction of continuity). That is to say, in the case in which a
pixel value f (0) in the origin (0, 0) in the pixel-of-interest
coordinates system in FIG. 261 (center of the pixel 3101 in FIG.
260) is taken as a pixel value f1, the direction where the pixel
value f1 continues is not the Y direction but the direction of data
continuity represented with the gradient G.sub.f (angle .theta.
direction).
In other words, in the case of conceiving the waveform of the
approximation function f(x) in a predetermined position y in the Y
direction (however, y is a numeric value other than zero), the
position corresponding to the pixel value f1 is not a position (0,
y) but a position (Cx (y), y) obtained by moving in the X direction
from the position (0, y) by a predetermined amount (here, let us
say that such an amount is also referred to as a shift amount.
Also, a shift amount is an amount depending on the position y in
the Y direction, so let us say that this shift amount is described
as C.sub.x (y)).
Accordingly, as the integral range of the right side of each of the
above Expression (174) through Expression (177), the integral range
needs to be set in light of the position y in the Y direction where
the center of the pixel value M (l) to be obtained (however, l is
any integer value of 1 through 4) exists, i.e., the shift amount
C.sub.x (Y).
Specifically, for example, the position y in the Y direction where
the centers of the pixel 3111 and pixel 3112 exist is not y=0 but
y=0.25.
Accordingly, the waveform of the approximation function f(x) at
y=0.25 is equivalent to a waveform obtained by moving the waveform
of the approximation function f(x) at y=0 by the shift amount
C.sub.x (0.25) in the X direction.
In other words, in the above Expression (174), if we say that the
pixel value M (1) as to the pixel 3111 is obtained by integrating
the approximation function f(x) at y=0 with a predetermined
integral range (from the start position x.sub.s1 to the end
position x.sub.e1), the integral range thereof becomes not a range
from the start position x.sub.s1=-0.5 to the end position
x.sub.e1=0 (a range itself where the pixel 3111 occupies in the X
direction) but the range shown in FIG. 261, i.e., from the start
position x.sub.s1=-0.5+C.sub.x (0.25) to the end position
x.sub.e1=0+C.sub.x (0.25) (a range where the pixel 3111 occupies in
the X direction in the event that the pixel 3111 is tentatively
moved by the shift amount C.sub.x (0.25)).
Similarly, in the above Expression (175), if we say that the pixel
value M (2) as to the pixel 3112 is obtained by integrating the
approximation function f(x) at y=0 with a predetermined integral
range (from the start position x.sub.s2 to the end position
x.sub.e2), the integral range thereof becomes not a range from the
start position x.sub.s2=0 to the end position x.sub.e2=0.5 (a range
itself where the pixel 3112 occupies in the X direction) but the
range shown in FIG. 261, i.e., from the start position
x.sub.s2=0+C.sub.x (0.25) to the end position x.sub.e1=0.5+C.sub.x
(0.25) (a range where the pixel 3112 occupies in the X direction in
the event that the pixel 3112 is tentatively moved by the shift
amount C.sub.x (0.25)).
Also, for example, the position y in the Y direction where the
centers of the pixel 3113 and pixel 3114 exist is not y=0 but
y=-0.25.
Accordingly, the waveform of the approximation function f(x) at
y=-0.25 is equivalent to a waveform obtained by moving the waveform
of the approximation function f(x) at y=0 by the shift amount
C.sub.x (-0.25) in the X direction.
In other words, in the above Expression (176), if we say that the
pixel value M (3) as to the pixel 3113 is obtained by integrating
the approximation function f(x) at y=0 with a predetermined
integral range (from the start position x.sub.s3 to the end
position x.sub.e3), the integral range thereof becomes not a range
from the start position x.sub.s3=-0.5 to the end position
x.sub.e3=0 (a range itself where the pixel 3113 occupies in the X
direction) but the range shown in FIG. 261, i.e., from the start
position x.sub.s3=-0.5+C.sub.x (-0.25) to the end position
x.sub.e3=0+C.sub.x (-0.25) (a range where the pixel 3113 occupies
in the X direction in the event that the pixel 3113 is tentatively
moved by the shift amount C.sub.x (-0.25)).
Similarly, in the above Expression (177), if we say that the pixel
value M (4) as to the pixel 3114 is obtained by integrating the
approximation function f(x) at y=0 with a predetermined integral
range (from the start position x.sub.s4 to the end position
x.sub.e4), the integral range thereof becomes not a range from the
start position x.sub.s4=0 to the end position x.sub.e4=0.5 (a range
itself where the pixel 3114 occupies in the X direction) but the
range shown in FIG. 261, i.e., from the start position
x.sub.s4=0+C.sub.x (-0.25) to the end position x.sub.e1=0.5+C.sub.x
(-0.25) (a range where the pixel 3114 occupies in the X direction
in the event that the pixel 3114 is tentatively moved by the shift
amount C.sub.x (-0.25)).
Accordingly, the image generating unit 102 (FIG. 259) calculates
the above Expression (174) through Expression (177) by substituting
the corresponding integral range of the above integral ranges for
each of these expressions, and outputs the calculated results of
these as the output pixel values M (1) through M (4).
Thus, the image generating unit 102 can create four pixels having
higher spatial resolution than that of the output pixel 3101, i.e.,
the pixel 3111 through pixel 3114 (FIG. 261) by employing the
one-dimensional reintegration method as a pixel at the output pixel
3101 (FIG. 260) from the sensor 2 (FIG. 259). Further, though not
shown in the drawing, as described above, the image generating unit
102 can create a pixel having an arbitrary powered spatial
resolution as to the output pixel 3101 without deterioration by
appropriately changing an integral range, in addition to the pixel
3111 through pixel 3114.
FIG. 262 represents a configuration example of the image generating
unit 103 employing such a one-dimensional reintegration method.
As shown in FIG. 262, the image generating unit 103 shown in this
example includes a conditions setting unit 3121, features storage
unit 3122, integral component calculation unit 3123, and output
pixel value calculation unit 3124.
The conditions setting unit 3121 sets the number of dimensions n of
the approximation function f(x) based on the actual world
estimating information (the features of the approximation function
f(x) in the example in FIG. 262) supplied from the actual world
estimating unit 102.
The conditions setting unit 3121 also sets an integral range in the
case of reintegrating the approximation function f(x) (in the case
of calculating an output pixel value). Note that an integral range
set by the conditions setting unit 3121 does not need to be the
width of a pixel. For example, the approximation function f(x) is
integrated in the spatial direction (X direction), and accordingly,
a specific integral range can be determined as long as the relative
size (power of spatial resolution) of an output pixel (pixel to be
calculated by the image generating unit 103) as to the spatial size
of each pixel of an input image from the sensor 2 (FIG. 259) is
known. Accordingly, the conditions setting unit 3121 can set, for
example, a spatial resolution power as an integral range.
The features storage unit 3122 temporally stores the features of
the approximation function f(x) sequentially supplied from the
actual world estimating unit 102. Subsequently, upon the features
storage unit 3122 storing all of the features of the approximation
function f(x), the features storage unit 3122 generates a features
table including all of the features of the approximation function
f(x), and supplies this to the output pixel value calculation unit
3124.
Incidentally, as described above, the image generating unit 103
calculates the output pixel value M using the above Expression
(172), but the approximation function f(x) included in the right
side of the above Expression (172) is represented as the following
Expression (178) specifically.
.function..times..times..times.d ##EQU00115##
Note that in Expression (178), w.sub.i represents the features of
the approximation function f(x) supplied from the actual world
estimating unit 102.
Accordingly, upon the approximation function f(x) of Expression
(178) being substituted for the approximation function f(x) of the
right side of the above Expression (172) so as to expand
(calculate) the right side of Expression (172), the output pixel
value M is represented as the following Expression (179).
.times..times..times..times..times..times..times..function.
##EQU00116##
In Expression (179), K.sub.i (x.sub.s, x.sub.e) represent the
integral components of the i-dimensional term. That is to say, the
integral components K.sub.i (x.sub.s, x.sub.e) are such as shown in
the following Expression (180).
.function..times. ##EQU00117##
The integral component calculation unit 3123 calculates the
integral components K.sub.i (x.sub.s, x.sub.e).
Specifically, as shown in Expression (180), the components K.sub.i
(x.sub.s, x.sub.e) can be calculated as long as the start position
x.sub.s and end position x.sub.e of an integral range, gain
G.sub.e, and i of the i-dimensional term are known.
Of these, the gain G.sub.e is determined with the spatial
resolution power (integral range) set by the conditions setting
unit 3121.
The range of i is determined with the number of dimensions n set by
the conditions setting unit 3121.
Also, each of the start position x.sub.s and end position x.sub.e
of an integral range is determined with the center pixel position
(x, y) and pixel width of an output pixel to be generated from now,
and the shift amount C.sub.x (y) representing the direction of data
continuity. Note that (x, y) represents the relative position from
the center position of a pixel of interest when the actual world
estimating unit 102 generates the approximation function f(x).
Further, each of the center pixel position (x, y) and pixel width
of an output pixel to be generated from now is determined with the
spatial resolution power (integral range) set by the conditions
setting unit 3121.
Also, with the shift amount C.sub.x (y), and the angle .theta.
supplied from the data continuity detecting unit 101, the relation
such as the following Expression (181) and Expression (182) holds,
and accordingly, the shift amount C.sub.x (y) is determined with
the angle .theta..
.times..times..theta.dd.function. ##EQU00118##
Note that in Expression (181), G.sub.f represents a gradient
representing the direction of data continuity, .theta. represents
an angle (angle generated between the X direction serving as one
direction of the spatial directions and the direction of data
continuity represented with a gradient G.sub.f) of one of the data
continuity information output from the data continuity detecting
unit 101 (FIG. 259). Also, dx represents the amount of fine
movement in the X direction, and dy represents the amount of fine
movement in the Y direction (spatial direction perpendicular to the
X direction) as to the dx.
Accordingly, the integral component calculation unit 3123
calculates the integral components K.sub.i (x.sub.s, x.sub.e) based
on the number of dimensions and spatial resolution power (integral
range) set by the conditions setting unit 3121, and the angle
.theta. of the data continuity information output from the data
continuity detecting unit 101, and supplies the calculated results
to the output pixel value calculation unit 3124 as an integral
component table.
The output pixel value calculation unit 3124 calculates the right
side of the above Expression (179) using the features table
supplied from the features storage unit 3122 and the integral
component table supplied from the integral component calculation
unit 3123, and outputs the calculation result as an output pixel
value M.
Next, description will be made regarding image generating
processing (processing in step S103 in FIG. 40) by the image
generating unit 103 (FIG. 262) employing the one-dimensional
reintegration method with reference to the flowchart in FIG.
263.
For example, now, let us say that the actual world estimating unit
102 has already generated the approximation function f(x) such as
shown in FIG. 260 while taking the pixel 3101 such as shown in FIG.
260 described above as a pixel of interest at the processing in
step S102 in FIG. 40 described above.
Also, let us say that the data continuity detecting unit 101 has
already output the angle .theta. such as shown in FIG. 260 as data
continuity information at the processing in step S101 in FIG. 40
described above.
In this case, the conditions setting unit 3121 sets conditions (the
number of dimensions and an integral range) at step S3101 in FIG.
263.
For example, now, let us say that 5 has been set as the number of
dimensions, and also a spatial quadruple density (spatial
resolution power to cause the pitch width of a pixel to become half
power in the upper/lower/left/right sides) has been set as an
integral range.
That is to say, in this case, consequently, it has been set that
the four pixel 3111 through pixel 3114 are created newly in a range
of -0.5 through 0.5 in the X direction, and also a range of -0.5
through 0.5 in the Y direction (in the range of the pixel 3101 in
FIG. 260), such as shown in FIG. 261.
In step S3102, the features storage unit 3122 acquires the features
of the approximation function f(x) supplied from the actual world
estimating unit 102, and generates a features table. In this case,
coefficients w.sub.0 through W.sub.5 of the approximation function
f(x) serving as a five-dimensional polynomial are supplied from the
actual world estimating unit 102, and accordingly, (w.sub.0,
w.sub.1, w.sub.2, w.sub.3, w.sub.4, w.sub.5) is generated as a
features table.
In step S3103, the integral component calculation unit 3123
calculates integral components based on the conditions (the number
of dimensions and integral range) set by the conditions setting
unit 3121, and the data continuity information (angle .theta.)
supplied from the data continuity detecting unit 101, and generates
an integral component table.
Specifically, for example, if we say that the respective pixels
3111 through 3114, which are to be generated from now, are appended
with numbers (hereafter, such a number is referred to as a mode
number) 1 through 4, the integral component calculation unit 3123
calculates the integral components K.sub.i (x.sub.s, x.sub.e) of
the above Expression (180) as a function of l (however, l
represents a mode number) such as integral components K.sub.i (l)
shown in the left side of the following Expression (183).
K.sub.i(l)=K.sub.i(x.sub.s,x.sub.e) (183)
Specifically, in this case, the integral components K.sub.i (l)
shown in the following Expression (184) are calculated.
.function..function..function..function..times..times..function..function-
..function..function..times..times..function..function..function..function-
..times..times..function..function..function..function.
##EQU00119##
Note that in Expression (184), the left side represents the
integral components K.sub.i (l), and the right side represents the
integral components K.sub.i (x.sub.s, x.sub.e). That is to say, in
this case, l is any one of 1 through 4, and also i is any one of 0
through 5, and accordingly, 24 K.sub.i (l) in total of 6 K.sub.i
(1), 6 K.sub.i (2), 6 K.sub.i (3), and 6 K.sub.i (4) are
calculated.
More specifically, first, the integral component calculation unit
3123 calculates each of the shift amounts C.sub.x (-0.25) and
C.sub.x (0.25) from the above Expression (181) and Expression (182)
using the angle .theta. supplied from the data continuity detecting
unit 101.
Next, the integral component calculation unit 3123 calculates the
integral components K.sub.i (x.sub.s, x.sub.e) of each right side
of the four expressions in Expression (184) regarding i=0 through 5
using the calculated shift amounts C.sub.x (-0.25) and C.sub.x
(0.25). Note that with this calculation of the integral components
K.sub.i (x.sub.s, x.sub.e), the above Expression (180) is
employed.
Subsequently, the integral component calculation unit 3123 converts
each of the 24 integral components K.sub.i (x.sub.s, x.sub.e)
calculated into the corresponding integral components K.sub.i (l)
in accordance with Expression (184), and generates an integral
component table including the 24 integral components K.sub.i (l)
converted (i.e., 6 K.sub.i (1), 6 K.sub.i (2), 6 K.sub.i (3), and 6
K.sub.i (4)).
Note that the sequence of the processing in step S3102 and the
processing in step S3103 is not restricted to the example in FIG.
263, the processing in step S3103 may be executed first, or the
processing in step S3102 and the processing in step S3103 may be
executed simultaneously.
Next, in step S3104, the output pixel value calculation unit 3124
calculates the output pixel values M (1) through M (4) respectively
based on the features table generated by the features storage unit
3122 at the processing in step S3102, and the integral component
table generated by the integral component calculation unit 3123 at
the processing in step S3103.
Specifically, in this case, the output pixel value calculation unit
3124 calculates each of the pixel value M (1) of the pixel 3111
(pixel of mode number 1), the pixel value M (2) of the pixel 3112
(pixel of mode number 2), the pixel value M (3) of the pixel 3113
(pixel of mode number 3), and the pixel value M (4) of the pixel
3114 (pixel of mode number 4) by calculating the right sides of the
following Expression (185) through Expression (188) corresponding
to the above Expression (179).
.function..times..times..function..function..times..times..function..func-
tion..times..times..function..function..times..times..function.
##EQU00120##
In step S3105, the output pixel value calculation unit 3124
determines regarding whether or not the processing of all the
pixels has been completed.
In step S3105, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S3102, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S3102 through S3104 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S3105, in the event that determination is made
that the processing of all the pixels has been completed), the
output pixel value calculation unit 3124 outputs the image in step
S3106. Then, the image generating processing ends.
Next, description will be made regarding the differences between
the output image obtained by employing the one-dimensional
reintegration method and the output image obtained by employing
another method (conventional classification adaptive processing)
regarding a predetermined input image with reference to FIG. 264
through FIG. 271.
FIG. 264 is a diagram illustrating the original image of the input
image, and FIG. 265 illustrates image data corresponding to the
original image in FIG. 264. In FIG. 265, the axis in the vertical
direction in the drawing represents pixel values, and the axis in
the lower right direction in the drawing represents the X direction
serving as one direction of the spatial directions of the image,
and the axis in the upper right direction in the drawing represents
the Y direction serving as the other direction of the spatial
directions of the image. Note that the respective axes in
later-described FIG. 267, FIG. 269, and FIG. 271 corresponds to the
axes in FIG. 265.
FIG. 266 is a diagram illustrating an example of an input image.
The input image illustrated in FIG. 266 is an image generated by
taking the mean of the pixel values of the pixels belonged to a
block made up of 2.times.2 pixels shown in FIG. 264 as the pixel
value of one pixel. That is to say, the input image is an image
obtained by integrating the image shown in FIG. 264 in the spatial
direction, which imitates the integration property of a sensor.
Also, FIG. 267 illustrates image data corresponding to the input
image in FIG. 266.
The original image illustrated in FIG. 264 includes a fine-line
image inclined almost 5.degree. clockwise from the vertical
direction. Similarly, the input image illustrated in FIG. 266
includes a fine-line image inclined almost 5.degree. clockwise from
the vertical direction.
FIG. 268 is a diagram illustrating an image (hereafter, the image
illustrated in FIG. 268 is referred to as a conventional image)
obtained by subjecting the input image illustrated in FIG. 266 to
conventional classification adaptive processing. Also, FIG. 269
illustrates image data corresponding to the conventional image.
Note that the classification adaptive processing is made up of
classification processing and adaptive processing, data is
classified based on the property thereof by the class
classification processing, and is subjected to the adaptive
processing for each class. With the adaptive processing, for
example, a low-quality or standard-quality image is subjected to
mapping using a predetermined tap coefficient so as to be converted
into a high-quality image.
FIG. 270 is a diagram illustrating an image (hereafter, the image
illustrated in FIG. 270 is referred to as an image according to the
present invention) obtained by applying the one-dimensional
reintegration method to which the present invention is applied, to
the input image illustrated in FIG. 266. Also, FIG. 271 illustrates
image data corresponding to the image according to the present
invention.
It can be understood that upon the conventional image in FIG. 268
being compared with the image according to the present invention in
FIG. 270, a fine-line image is different from that in the original
image in FIG. 264 in the conventional image, but on the other hand,
the fine-line image is almost the same as that in the original
image in FIG. 264 in the image according to the present
invention.
This difference is caused by a difference wherein the conventional
class classification adaptation processing is a method for
performing processing on the basis (origin) of the input image in
FIG. 266, but on the other hand, the one-dimensional reintegration
method according to the present invention is a method for
estimating the original image in FIG. 264 (generating the
approximation function f(x) corresponding to the original image) in
light of continuity of a fine line, and performing processing
(performing reintegration so as to calculate pixel values) on the
basis (origin) of the original image estimated.
Thus, with the one-dimensional reintegration method, an output
image (pixel values) is generated by integrating the approximation
function f(x) in an arbitrary range on the basis (origin) of the
approximation function f(x) (the approximation function f(x) of the
X cross-sectional waveform F(x) in the actual world) serving as the
one-dimensional polynomial generated with the one-dimensional
polynomial approximating method.
Accordingly, with the one-dimensional reintegration method, it
becomes possible to output an image more similar to the original
image (the light signal in the actual world 1 which is to be cast
in the sensor 2) in comparison with the conventional other
methods.
In other words, the one-dimensional reintegration method is based
on condition that the data continuity detecting unit 101 in FIG.
259 detects continuity of data in an input image made up of
multiple pixels having a pixel value on which the light signals in
the actual world 1 are projected by the multiple detecting elements
of the sensor 2 each having spatio-temporal integration effects,
and projected by the detecting elements of which a part of
continuity of the light signals in the actual world 1 drops, and in
response to the detected continuity of data, the actual world
estimating unit 102 estimates the light signal function F by
approximating the light signal function F (specifically, X
cross-sectional waveform F(x)) representing the light signals in
the actual world 1 with a predetermined approximation function f(x)
on assumption that the pixel value of a pixel corresponding to a
position in the one-dimensional direction of the time-space
directions of the input image is the pixel value acquired by
integration effects in the one-dimensional direction thereof.
Speaking in detail, for example, the one-dimensional reintegration
method is based on condition that the X cross-sectional waveform
F(x) is approximated with the approximation function f(x) on
assumption that the pixel value of each pixel corresponding to a
distance along in the one-dimensional direction from a line
corresponding to the detected continuity of data is the pixel value
obtained by the integration effects in the one-dimensional
direction thereof.
With the one-dimensional reintegration method, for example, the
image generating unit 103 in FIG. 259 (FIG. 3) generates a pixel
value M corresponding to a pixel having a desired size by
integrating the X cross-sectional waveform F(x) estimated by the
actual world estimating unit 102, i.e., the approximation function
f(x) in desired increments in the one-dimensional direction based
on such an assumption, and outputs this as an output image.
Accordingly, with the one-dimensional reintegration method, it
becomes possible to output an image more similar to the original
image (the light signal in the actual world 1 which is to be cast
in the sensor 2) in comparison with the conventional other
methods.
Also, with the one-dimensional reintegration method, as described
above, the integral range is arbitrary, and accordingly, it becomes
possible to create resolution (temporal resolution or spatial
resolution) different from the resolution of an input image by
varying the integral range. That is to say, it becomes possible to
generate an image having arbitrary powered resolution as well as an
integer value as to the resolution of the input image.
Further, the one-dimensional reintegration method enables
calculation of an output image (pixel values) with less calculation
processing amount than other reintegration methods.
Next, description will be made regarding a two-dimensional
reintegration method with reference to FIG. 272 through FIG.
278.
The two-dimensional reintegration method is based on condition that
the approximation function f(x, y) has been generated with the
two-dimensional polynomial approximating method.
That is to say, for example, it is an assumption that the image
function F(x, y, t) representing the light signal in the actual
world 1 (FIG. 259) having continuity in the spatial direction
represented with the gradient G.sub.F has been approximated with a
waveform projected in the spatial directions (X direction and Y
direction), i.e., the waveform F(x, y) on the X-Y plane has been
approximated with the approximation function f(x, y) serving as a
n-dimensional (n is an arbitrary integer) polynomial, such as shown
in FIG. 272.
In FIG. 272, the horizontal direction represents the X direction
serving as one direction in the spatial directions, the upper right
direction represents the Y direction serving as the other direction
in the spatial directions, and the vertical direction represents
light levels, respectively in the drawing. G.sub.F represents
gradient as continuity in the spatial directions.
Note that with the example in FIG. 272, the direction of continuity
is taken as the spatial directions (X direction and Y direction),
so the projection function of a light signal to be approximated is
taken as the function F(x, y), but as described later, the function
F(x, t) or function F(y, t) may be a target of approximation
according to the direction of continuity.
In the case of the example in FIG. 272, with the two-dimensional
reintegration method, the output pixel value M is calculated as the
following Expression (189).
.times..intg..times..intg..times..function..times.d.times.d
##EQU00121##
Note that in Expression (189), y.sub.s represents an integration
start position in the Y direction, and y.sub.e represents an
integration end position in the Y direction. Similarly, x.sub.s
represents an integration start position in the X direction, and
x.sub.e represents an integration end position in the X direction.
Also, G.sub.e represents a predetermined gain.
In Expression (189), an integral range can be set arbitrarily, and
accordingly, with the two-dimensional reintegration method, it
becomes possible to create pixels having an arbitrary powered
spatial resolution as to the original pixels (the pixels of an
input image from the sensor 2 (FIG. 259)) without deterioration by
appropriately changing this integral range.
FIG. 273 represents a configuration example of the image generating
unit 103 employing the two-dimensional reintegration method.
As shown in FIG. 273, the image generating unit 103 in this example
includes a conditions setting unit 3201, features storage unit
3202, integral component calculation unit 3203, and output pixel
value calculation unit 3204.
The conditions setting unit 3201 sets the number of dimensions n of
the approximation function f(x, y) based on the actual world
estimating information (with the example in FIG. 273, the features
of the approximation function f(x, y)) supplied from the actual
world estimating unit 102.
The conditions setting unit 3201 also sets an integral range in the
case of reintegrating the approximation function f(x, y) (in the
case of calculating an output pixel value). Note that an integral
range set by the conditions setting unit 3201 does not need to be
the vertical width or the horizontal width of a pixel. For example,
the approximation function f(x, y) is integrated in the spatial
directions (X direction and Y direction), and accordingly, a
specific integral range can be determined as long as the relative
size (power of spatial resolution) of an output pixel (pixel to be
generated from now by the image generating unit 103) as to the
spatial size of each pixel of an input image from the sensor 2 is
known. Accordingly, the conditions setting unit 3201 can set, for
example, a spatial resolution power as an integral range.
The features storage unit 3202 temporally stores the features of
the approximation function f(x, y) sequentially supplied from the
actual world estimating unit 102. Subsequently, upon the features
storage unit 3202 storing all of the features of the approximation
function f(x, y), the features storage unit 3202 generates a
features table including all of the features of the approximation
function f(x, y), and supplies this to the output pixel value
calculation unit 3204.
Now, description will be made regarding the details of the
approximation function f(x, y).
For example, now, let us say that the light signals (light signals
represented with the wave F (x, y)) in the actual world 1 (FIG.
259) having continuity in the spatial directions represented with
the gradient G.sub.F shown in FIG. 272 described above have been
detected by the sensor 2 (FIG. 259), and have been output as an
input image (pixel values).
Further, for example, let us say that the data continuity detecting
unit 101 (FIG. 3) has subjected a region 3221 of an input image
made up of 20 pixels in total (20 squares represented with a dashed
line in the drawing) of 4 pixels in the X direction and also 5
pixels in the Y direction of this input image to the processing
thereof, and has output an angle .theta. (angle .theta. generated
between the direction of data continuity represented with the
gradient G.sub.f corresponding to the gradient G.sub.F and the X
direction) as one of data continuity information, as shown in FIG.
274.
Note that as viewed from the actual world estimating unit 102, the
data continuity detecting unit 101 should simply output the angle
.theta. at a pixel of interest, and accordingly, the processing
region of the data continuity detecting unit 101 is not restricted
to the above region 3221 in the input image.
Also, with the region 3221 in the input image, the horizontal
direction in the drawing represents the X direction serving as one
direction of the spatial directions, and the vertical direction in
the drawing represents the Y direction serving the other direction
of the spatial directions.
Further, in FIG. 274, a pixel, which is the second pixel from the
left, and also the third pixel from the bottom, is taken as a pixel
of interest, and an (x, y) coordinates system is set so as to take
the center of the pixel of interest as the origin (0, 0). A
relative distance (hereafter, referred to as a cross-sectional
direction distance) in the X direction as to a straight line
(straight line of the gradient G.sub.f representing the direction
of data continuity) having an angle .theta. passing through the
origin (0, 0) is taken as x'.
Further, in FIG. 274, the graph on the right side represents the
approximation function f(x') serving as a n-dimensional (n is an
arbitrary integer) polynomial, which is a function approximating a
one-dimensional waveform (hereafter, referred to as an X
cross-sectional waveform F(x')) wherein the image function F(x, y,
t) of which variables are positions x, y, and z on the
three-dimensional space, and point-in-time t is projected in the X
direction at an arbitrary position y in the Y direction. Of the
axes in the graph on the right side, the axis in the horizontal
direction in the drawing represents a cross-sectional direction
distance, and the axis in the vertical direction in the drawing
represents pixel values.
In this case, the approximation function f(x') shown in FIG. 274 is
a n-dimensional polynomial, so is represented as the following
Expression (190).
.function.'.times.'.times.'.times.'.times..times..times..times.'.times..t-
imes. ##EQU00122##
Also, since the angle .theta. is determined, the straight line
having angle .theta. passing through the origin (0, 0) is uniquely
determined, and a position x.sub.1 in the X direction of the
straight line at an arbitrary position y in the Y direction is
represented as the following Expression (191). However, in
Expression (191), s represents cot .theta.. x.sub.1=s.times.y
(191)
That is to say, as shown in FIG. 274, a point on the straight line
corresponding to continuity of data represented with the gradient
G.sub.f is represented with a coordinate value (x.sub.1, y).
The cross-sectional direction distance x' is represented as the
following Expression (192) using Expression (191).
x'=x-x.sub.1=x-s.times.y (192)
Accordingly, the approximation function f(x, y) at an arbitrary
position (x, y) within the input image region 3221 is represented
as the following Expression (193) using Expression (190) and
Expression (192).
.function..times..function..times. ##EQU00123##
Note that in Expression (193), w.sub.i represents the features of
the approximation function f(x, y).
Now, description will return to FIG. 273, wherein the features
w.sub.i included in Expression (193) are supplied from the actual
world estimating unit 102, and stored in the features storage unit
3202. Upon the features storage unit 3202 storing all of the
features w.sub.i represented with Expression (193), the features
storage unit 3202 generates a features table including all of the
features w.sub.i, and supplies this to the output pixel value
calculation unit 3204.
Also, upon the right side of the above Expression (189) being
expanded (calculated) by substituting the approximation function
f(x, y) of Expression (193) for the approximation function f(x, y)
in the right side of Expression (189), the output pixel value M is
represented as the following Expression (194).
.times..times..times..times..times..times..times..times..function..times.-
.times..times..times..function. ##EQU00124##
In Expression (194), K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e)
represent the integral components of the i-dimensional term. That
is to say, the integral components K.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e) are such as shown in the following Expression
(195).
.function..times..times..times..times..times..times..function..times.
##EQU00125##
The integral component calculation unit 3203 calculates the
integral components K.sub.i (x.sub.s, x.sub.e, y.sub.s,
y.sub.e).
Specifically, as shown in Expression (194) and Expression (195),
the integral components K.sub.i (x.sub.s, x.sub.e, y.sub.s,
y.sub.e) can be calculated as long as the start position x.sub.s in
the X direction and end position x.sub.e in the X direction of an
integral range, the start position y.sub.s in the Y direction and
end position y.sub.e in the Y direction of an integral range,
variable s, gain G.sub.e, and i of the i-dimensional term are
known.
Of these, the gain G.sub.e is determined with the spatial
resolution power (integral range) set by the conditions setting
unit 3201.
The range of i is determined with the number of dimensions n set by
the conditions setting unit 3201.
A variable s is, as described above, cot .theta., so is determined
with the angle .theta. output from the data continuity detecting
unit 101.
Also, each of the start position x.sub.s in the X direction and end
position x.sub.e in the X direction of an integral range, and the
start position y.sub.s in the Y direction and end position y.sub.e
in the Y direction of an integral range is determined with the
center pixel position (x, y) and pixel width of an output pixel to
be generated from now. Note that (x, y) represents a relative
position from the center position of the pixel of interest when the
actual world estimating unit 102 generates the approximation
function f(x).
Further, each of the center pixel position (x, y) and pixel width
of an output pixel to be generated from now is determined with the
spatial resolution power (integral range) set by the conditions
setting unit 3201.
Accordingly, the integral component calculation unit 3203
calculates K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e) based on
the number of dimensions and the spatial resolution power (integral
range) set by the conditions setting unit 3201, and the angle
.theta. of the data continuity information output from the data
continuity detecting unit 101, and supplies the calculated result
to the output pixel value calculation unit 3204 as an integral
component table.
The output pixel value calculation unit 3204 calculates the right
side of the above Expression (194) using the features table
supplied from the features storage unit 3202, and the integral
component table supplied from the integral component calculation
unit 3203, and outputs the calculated result to the outside as the
output pixel value M.
Next, description will be made regarding image generating
processing (processing in step S103 in FIG. 40) by the image
generating unit 103 (FIG. 274) employing the two-dimensional
reintegration method with reference to the flowchart in FIG.
275.
For example, let us say that the light signals represented with the
function F(x, y) shown in FIG. 272 have been cast in the sensor 2
so as to become an input image, and the actual world estimating
unit 102 has already generated the approximation function f(x, y)
for approximating the function F(x, y) with one pixel 3231 such as
shown in FIG. 276 as a pixel of interest at the processing in step
S102 in FIG. 40 described above.
Note that in FIG. 276, the pixel value (input pixel value) of the
pixel 3231 is taken as P, and the shape of the pixel 3231 is taken
as a square of which one side is 1 in length. Also, of the spatial
directions, the direction in parallel with one side of the pixel
3231 is taken as the X direction, and the direction orthogonal to
the X direction is taken as the Y direction. Further, a coordinates
system (hereafter, referred to as a pixel-of-interest coordinates
system) in the spatial directions (X direction and Y direction) of
which the origin is the center of the pixel 3231 is set.
Also, let us say that in FIG. 276, the data continuity detecting
unit 101, which takes the pixel 3231 as a pixel of interest, has
already output the angle .theta. as data continuity information
corresponding to continuity of data represented with the gradient
G.sub.f at the processing in step S101 in FIG. 40 described
above.
Description will return to FIG. 275, and in this case, the
conditions setting unit 3201 sets conditions (the number of
dimensions and an integral range) at step S3201.
For example, now, let us say that 5 has been set as the number of
dimensions, and also spatial quadruple density (spatial resolution
power to cause the pitch width of a pixel to become half power in
the upper/lower/left/right sides) has been set as an integral
range.
That is to say, in this case, it has been set that the four pixel
3241 through pixel 3244 are created newly in a range of -0.5
through 0.5 in the X direction, and also a range of -0.5 through
0.5 in the Y direction (in the range of the pixel 3231 in FIG.
276), such as shown in FIG. 277. Note that in FIG. 277 as well, the
same pixel-of-interest coordinates system as that in FIG. 276 is
shown.
Also, in FIG. 277, M (1) represents the pixel value of the pixel
3241 to be generated from now, M (2) represents the pixel value of
the pixel 3242 to be generated from now, M (3) represents the pixel
value of the pixel 3243 to be generated from now, and M (4)
represents the pixel value of the pixel 3244 to be generated from
now.
Description will return to FIG. 275, in step S3202, the features
storage unit 3202 acquires the features of the approximation
function f(x, y) supplied from the actual world estimating unit
102, and generates a features table. In this case, the coefficients
w.sub.0 through w.sub.5 of the approximation function f(x) serving
as a 5-dimensional polynomial are supplied from the actual world
estimating unit 102, and accordingly, (w.sub.0, w.sub.1, w.sub.2,
w.sub.3, w.sub.4, w.sub.5) is generated as a features table.
In step S3203, the integral component calculation unit 3203
calculates integral components based on the conditions (the number
of dimensions and an integral range) set by the conditions setting
unit 3201, and the data continuity information (angle .theta.)
supplied from the data continuity detecting unit 101, and generates
an integral component table.
Specifically, for example, let us say that numbers (hereafter, such
a number is referred to as a mode number) 1 through 4 are
respectively appended to the pixel 3241 through pixel 3244 to be
generated from now, the integral component calculation unit 3203
calculates the integral components K.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e) of the above Expression (194) as a function of l
(however, l represents a mode number) such as the integral
components K.sub.i (l) shown in the left side of the following
Expression (196).
K.sub.i(l)=K.sub.i(x.sub.s,x.sub.e,y.sub.s,y.sub.e) (196)
Specifically, in this case, the integral components K.sub.i (l)
shown in the following Expression (197) are calculated.
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function. ##EQU00126##
Note that in Expression (197), the left side represents the
integral components K.sub.i (l), and the right side represents the
integral components K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e).
That is to say, in this case, l is any one of 1 thorough 4, and
also i is any one of 0 through 5, and accordingly, 24 K.sub.i (l)
in total of 6 K.sub.i (1), 6 K.sub.i (2), 6 K.sub.i (3), and 6
K.sub.i (4) are calculated.
More specifically, first, the integral component calculation unit
3203 calculates the variable s (s=cot .theta.) of the above
Expression (191) using the angle .theta. supplied from the data
continuity detecting unit 101.
Next, the integral component calculation unit 3203 calculates the
integral components K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e) of
each right side of the four expressions in Expression (197)
regarding i=0 through 5 using the calculated variable s. Note that
with this calculation of the integral components K.sub.i (x.sub.s,
x.sub.e, y.sub.s, y.sub.e), the above Expression (194) is
employed.
Subsequently, the integral component calculation unit 3203 converts
each of the 24 integral components K.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e) calculated into the corresponding integral
components K.sub.i (l) in accordance with Expression (197), and
generates an integral component table including the 24 integral
components K.sub.i (l) converted (i.e., 6 K.sub.i (1), 6 K.sub.i
(2), 6 K.sub.i (3), and 6 K.sub.i (4)).
Note that the sequence of the processing in step S3202 and the
processing in step S3203 is not restricted to the example in FIG.
275, the processing in step S3203 may be executed first, or the
processing in step S3202 and the processing in step S3203 may be
executed simultaneously.
Next, in step S3204, the output pixel value calculation unit 3204
calculates the output pixel values M (1) through M (4) respectively
based on the features table generated by the features storage unit
3202 at the processing in step S3202, and the integral component
table generated by the integral component calculation unit 3203 at
the processing in step S3203.
Specifically, in this case, the output pixel value calculation unit
3204 calculates each of the pixel value M (1) of the pixel 3241
(pixel of mode number 1), the pixel value M (2) of the pixel 3242
(pixel of mode number 2), the pixel value M (3) of the pixel 3243
(pixel of mode number 3), and the pixel value M (4) of the pixel
3244 (pixel of mode number 4) shown in FIG. 254 by calculating the
right sides of the following Expression (198) through Expression
(201) corresponding to the above Expression (194).
.function..times..times..function..function..times..times..function..func-
tion..times..times..function..function..times..times..function.
##EQU00127##
However, in this case, each n of Expression (198) through
Expression (201) becomes 5.
In step S3205, the output pixel value calculation unit 3204
determines regarding whether or not the processing of all the
pixels has been completed.
In step S3205, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S3202, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S3202 through S3204 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S3205, in the event that determination is made
that the processing of all the pixels has been completed), the
output pixel value calculation unit 3204 outputs the image in step
S3206. Then, the image generating processing ends.
Thus, four pixels having higher spatial resolution than the input
pixel 3231, i.e., the pixel 3241 through pixel 3244 (FIG. 277) can
be created by employing the two-dimensional reintegration method as
a pixel at the pixel 3231 of the input image (FIG. 276) from the
sensor 2 (FIG. 259). Further, though not shown in the drawing, as
described above, the image generating unit 103 can create a pixel
having an arbitrary powered spatial resolution as to the input
pixel 3231 without deterioration by appropriately changing an
integral range, in addition to the pixel 3241 through pixel
3244.
As described above, as description of the two-dimensional
reintegration method, an example for subjecting the approximation
function f(x, y) as to the spatial directions (X direction and Y
direction) to two-dimensional integration has been employed, but
the two-dimensional reintegration method can be applied to the
time-space directions (X direction and t direction, or Y direction
and t direction).
That is to say, the above example is an example in the case in
which the light signals in the actual world 1 (FIG. 259) have
continuity in the spatial directions represented with the gradient
G.sub.F such as shown in FIG. 272, and accordingly, an expression
including two-dimensional integration in the spatial directions (X
direction and Y direction) such as shown in the above Expression
(189) has been employed. However, the concept regarding
two-dimensional integration can be applied not only to the spatial
direction but also the time-space directions (X direction and t
direction, or Y direction and t direction).
In other words, with the two-dimensional polynomial approximating
method serving as an assumption of the two-dimensional
reintegration method, it is possible to perform approximation using
a two-dimensional polynomial even in the case in which the image
function F(x, y, t) representing the light signals has continuity
in the time-space directions (however, X direction and t direction,
or Y direction and t direction) as well as continuity in the
spatial directions.
Specifically, for example, in the event that there is an object
moving horizontally in the X direction at uniform velocity, the
direction of movement of the object is represented with like a
gradient V.sub.F in the X-t plane such as shown in FIG. 278. In
other words, it can be said that the gradient V.sub.F represents
the direction of continuity in the time-space directions in the X-t
plane. Accordingly, the data continuity detecting unit 101 (FIG.
259) can output movement .theta. such as shown in FIG. 278
(strictly speaking, though not shown in the drawing, movement
.theta. is an angle generated by the direction of data continuity
represented with the gradient V.sub.f corresponding to the gradient
V.sub.F and the X direction in the spatial direction) as data
continuity information corresponding to the gradient V.sub.F
representing continuity in the time-space directions in the X-t
plane as well as the above angle .theta. (data continuity
information corresponding to the gradient G.sub.F representing
continuity in the spatial directions in the X-Y plane).
Also, the actual world estimating unit 102 (FIG. 259) employing the
two-dimensional polynomial approximating method can calculate the
coefficients (features) w.sub.i of an approximation function f(x,
t) with the same method as the above method by employing the
movement .theta. instead of the angle .theta.. However, in this
case, the equation to be employed is not the above Expression (193)
but the following Expression (202).
.function..times..function..times. ##EQU00128##
Note that in Expression (202), s is cot .theta. (however, .theta.
is movement).
Accordingly, the image generating unit 103 (FIG. 259) employing the
two-dimensional reintegration method can calculate the pixel value
M by substituting the f (x, t) of the above Expression (202) for
the right side of the following Expression (203), and calculating
this.
.times..intg..times..intg..times..function..times.d.times.d
##EQU00129##
Note that in Expression (203), t.sub.s represents an integration
start position in the t direction, and t.sub.e represents an
integration end position in the t direction. Similarly, x.sub.s
represents an integration start position in the X direction, and
x.sub.e represents an integration end position in the X direction.
G.sub.e represents a predetermined gain.
Alternately, an approximation function f(y, t) focusing attention
on the spatial direction Y instead of the spatial direction X can
be handled as the same way as the above approximation function f(x,
t).
Incidentally, in Expression (202), it becomes possible to obtain
data not integrated in the temporal direction, i.e., data without
movement blurring by regarding the t direction as constant, i.e.,
by performing integration while ignoring integration in the t
direction. In other words, this method may be regarded as one of
two-dimensional reintegration methods in that reintegration is
performed on condition that one certain dimension of
two-dimensional polynomials is constant, or in fact, may be
regarded as one of one-dimensional reintegration methods in that
one-dimensional reintegration in the X direction is performed.
Also, in Expression (203), an integral range may be set
arbitrarily, and accordingly, with the two-dimensional
reintegration method, it becomes possible to create a pixel having
an arbitrary powered resolution as to the original pixel (pixel of
an input image from the sensor 2 (FIG. 259)) without deterioration
by appropriately changing this integral range.
That is to say, with the two-dimensional reintegration method, it
becomes possible to create temporal resolution by appropriately
changing an integral range in the temporal direction t. Also, it
becomes possible to create spatial resolution by appropriately
changing an integral range in the spatial direction X (or spatial
direction Y). Further, it becomes possible to create both temporal
resolution and spatial resolution by appropriately changing each
integral range in the temporal direction and in the spatial
direction X.
Note that as described above, creation of any one of temporal
resolution and spatial resolution may be performed even with the
one-dimensional reintegration method, but creation of both temporal
resolution and spatial resolution cannot be performed with the
one-dimensional reintegration method in theory, which becomes
possible only by performing two-dimensional or more reintegration.
That is to say, creation of both temporal resolution and spatial
resolution becomes possible only by employing the two-dimensional
reintegration method and a later-described three-dimensional
reintegration method.
Also, the two-dimensional reintegration method takes not
one-dimensional but two-dimensional integration effects into
consideration, and accordingly, an image more similar to the light
signal in the actual world 1 (FIG. 259) may be created.
In other words, with the two-dimensional reintegration method, for
example, the data continuity detecting unit 101 in FIG. 259 (FIG.
3) detects continuity (e.g., continuity of data represented with
the gradient G.sub.f in FIG. 274) of data in an input image made up
of multiple pixels having a pixel value on which the light signals
in the actual world 1 are projected by the multiple detecting
elements of the sensor 2 each having spatio-temporal integration
effects, and projected by the detecting elements of which a part of
continuity (e.g., continuity represented with the gradient G.sub.F
in FIG. 272) of the light signals in the actual world 1 drops.
Subsequently, for example, in response to the continuity of data
detected by the data continuity detecting unit 101, the actual
world estimating unit 102 in FIG. 259 (FIG. 3) estimates the light
signal function F by approximating the light signal function
F(specifically, function F(x, y) in FIG. 272) representing the
light signals in the actual world 1 with an approximation function
f(x, y), which is a polynomial, on assumption that the pixel value
of a pixel corresponding to at least a position in the
two-dimensional direction (e.g., spatial direction X and spatial
direction Y in FIG. 272) of the time-space directions of the image
data is the pixel value acquired by at least integration effects in
the two-dimensional direction, which is an assumption.
Speaking in detail, for example, the actual world estimating unit
102 estimates a first function representing the light signals in
the real world by approximating the first function with a second
function serving as a polynomial on condition that the pixel value
of a pixel corresponding to at least a distance (for example,
cross-sectional direction distance x' in FIG. 274) along in the
two-dimensional direction from a line corresponding to continuity
of data (for example, a line (arrow) corresponding to the gradient
G.sub.f in FIG. 274) detected by the continuity detecting unit 101
is the pixel value acquired by at least integration effects in the
two-dimensional direction, which is an assumption.
With the two-dimensional reintegration method, based on such an
assumption, for example, the image generating unit 103 (FIG. 273
for configuration) in FIG. 259 (FIG. 3) generates a pixel value
corresponding to a pixel (for example, output image (pixel value M)
in FIG. 259. Specifically, for example, the pixel 3241 through
pixel 3244 in FIG. 277) having a desired size by integrating the
function F(x, y) estimated by the actual world estimating unit 102,
i.e., the approximation function f(x, y) in at least desired
increments in the two-dimensional direction (e.g., by calculating
the right side of the above Expression (186)).
Accordingly, the two-dimensional reintegration method enables not
only any one of temporal resolution and spatial resolution but also
both temporal resolution and spatial resolution to be created.
Also, with the two-dimensional reintegration method, an image more
similar to the light signal in the actual world 1 (FIG. 259) than
that in the one-dimensional reintegration method may be
generated.
Next, description will be made regarding a three-dimensional
reintegration method with reference to FIG. 279 and FIG. 280.
With the three-dimensional reintegration method, the approximation
function f(x, y, t) has been created using the three-dimensional
function approximating method, which is an assumption.
In this case, with the three-dimensional reintegration method, the
output pixel value M is calculated as the following Expression
(204).
.times..intg..times..intg..times..intg..times..function..times.d.times.d.-
times.d ##EQU00130##
Note that in Expression (204), t.sub.s represents an integration
start position in the t direction, and t.sub.e represents an
integration end position in the t direction. Similarly, y.sub.s
represents an integration start position in the Y direction, and
y.sub.e represents an integration end position in the Y direction.
Also, x.sub.s represents an integration start position in the X
direction, and x.sub.e represents an integration end position in
the X direction. G.sub.e represents a predetermined gain.
Also, in Expression (204), an integral range may be set
arbitrarily, and accordingly, with the three-dimensional
reintegration method, it becomes possible to create a pixel having
an arbitrary powered time-space resolution as to the original pixel
(pixel of an input image from the sensor 2 (FIG. 259)) without
deterioration by appropriately changing this integral range. That
is to say, upon the integral range in the spatial direction being
reduced, a pixel pitch can be reduced without restraint. On the
other hand, upon the integral range in the spatial direction being
enlarged, a pixel pitch can be enlarged without restraint. Also,
upon the integral range in the temporal direction being reduced,
temporal resolution can be created based on an actual waveform.
FIG. 279 represents a configuration example of the image generating
unit 103 employing the three-dimensional reintegration method.
As shown in FIG. 279, this example of the image generating unit 103
includes a conditions setting unit 3301, features storage unit
3302, integral component calculation unit 3303, and output pixel
value calculation unit 3304.
The conditions setting unit 3301 sets the number of dimensions n of
the approximation function f(x, y, t) based on the actual world
estimating information (with the example in FIG. 279, features of
the approximation function f(x, y, t)) supplied from the actual
world estimating unit 102.
The conditions setting unit 3301 sets an integral range in the case
of reintegrating the approximation function f(x, y, t) (in the case
of calculating output pixel values). Note that an integral range
set by the conditions setting unit 3301 needs not to be the width
(vertical width and horizontal width) of a pixel or shutter time
itself. For example, it becomes possible to determine a specific
integral range in the spatial direction as long as the relative
size (spatial resolution power) of an output pixel (pixel to be
generated from now by the image generating unit 103) as to the
spatial size of each pixel of an input image from the sensor 2
(FIG. 259) is known. Similarly, it becomes possible to determine a
specific integral range in the temporal direction as long as the
relative time (temporal resolution power) of an output pixel as to
the shutter time of the sensor 2 (FIG. 259) is known. Accordingly,
the conditions setting unit 3301 can set, for example, a spatial
resolution power and temporal resolution power as an integral
range.
The features storage unit 3302 temporally stores the features of
the approximation function f(x, y, t) sequentially supplied from
the actual world estimating unit 102. Subsequently, upon the
features storage unit 3302 storing all of the features of the
approximation function f(x, y, t), the features storage unit 3302
generates a features table including all of the features of the
approximation function f(x, y, t), and supplies this to the output
pixel value calculation unit 3304.
Incidentally, upon the right side of the approximation function
f(x, y) of the right side of the above Expression (204) being
expanded (calculated), the output pixel value M is represented as
the following Expression (205).
.times..times..function. ##EQU00131##
In Expression (205), K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e,
t.sub.s, t.sub.e) represents the integral components of the
i-dimensional term. However, x.sub.s represents an integration
range start position in the X direction, x.sub.e represents an
integration range end position in the X direction, y.sub.s
represents an integration range start position in the Y direction,
y.sub.e represents an integration range end position in the Y
direction, t.sub.s represents an integration range start position
in the t direction, and t.sub.e represents an integration range end
position in the t direction, respectively.
The integral component calculation unit 3303 calculates the
integral components K.sub.i (x.sub.s, x.sub.e, y.sub.s, y.sub.e,
t.sub.s, t.sub.e).
Specifically, the integral component calculation unit 3303
calculates the integral components K.sub.i (x.sub.s, x.sub.e,
y.sub.s, y.sub.e, t.sub.s, t.sub.e) based on the number of
dimensions and the integral range (spatial resolution power or
temporal resolution power) set by the conditions setting unit 3301,
and the angle .theta. or movement .theta. of the data continuity
information output from the data continuity detecting unit 101, and
supplies the calculated results to the output pixel value
calculation unit 3304 as an integral component table.
The output pixel value calculation unit 3304 calculates the right
side of the above Expression (205) using the features table
supplied from the features storage unit 3302, and the integral
component table supplied from the integral component calculation
unit 3303, and outputs the calculated result to the outside as the
output pixel value M.
Next, description will be made regarding image generating
processing (processing in step S103 in FIG. 40) by the image
generating unit 103 (FIG. 279) employing the three-dimensional
reintegration method with reference to the flowchart in FIG.
280.
For example, let us say that the actual world estimating unit 102
(FIG. 259) has already generated an approximation function f(x, y,
t) for approximating the light signals in the actual world 1 (FIG.
259) with a predetermined pixel of an input image as a pixel of
interest at the processing in step S102 in FIG. 40 described
above.
Also, let us say that the data continuity detecting unit 101 (FIG.
259) has already output the angle .theta. or movement .theta. as
data continuity information with the same pixel as the actual world
estimating unit 102 as a pixel of interest.
In this case, the conditions setting unit 3301 sets conditions (the
number of dimensions and an integral range) at step S3301 in FIG.
280.
In step S3302, the features storage unit 3302 acquires the features
w.sub.i of the approximation function f(x, y, t) supplied from the
actual world estimating unit 102, and generates a features
table.
In step S3303, the integral component calculation unit 3303
calculates integral components based on the conditions (the number
of dimensions and an integral range) set by the conditions setting
unit 3301, and the data continuity information (angle .theta. or
movement .theta.) supplied from the data continuity detecting unit
101, and generates an integral component table.
Note that the sequence of the processing in step S3302 and the
processing in step S3303 is not restricted to the example in FIG.
280, the processing in step S3303 may be executed first, or the
processing in step S3302 and the processing in step S3303 may be
executed simultaneously.
Next, in step S3304, the output pixel value calculation unit 3304
calculates each output pixel value based on the features table
generated by the features storage unit 3302 at the processing in
step S3302, and the integral component table generated by the
integral component calculation unit 3303 at the processing in step
S3303.
In step S3305, the output pixel value calculation unit 3304
determines regarding whether or not the processing of all the
pixels has been completed.
In step S3305, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S3302, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S3302 through S3304 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S3305, in the event that determination is made
that the processing of all the pixels has been completed), the
output pixel value calculation unit 3304 outputs the image in step
S3306. Then, the image generating processing ends.
Thus, in the above Expression (204), an integral range may be set
arbitrarily, and accordingly, with the three-dimensional
reintegration method, it becomes possible to create a pixel having
an arbitrary powered resolution as to the original pixel (pixel of
an input image from the sensor 2 (FIG. 259)) without deterioration
by appropriately changing this integral range.
That is to say, with the three-dimensional reintegration method,
appropriately changing an integral range in the temporal direction
enables temporal resolution to be created. Also, appropriately
changing an integral range in the spatial direction enables spatial
resolution to be created. Further, appropriately changing each
integral range in the temporal direction and in the spatial
direction enables both temporal resolution and spatial resolution
to be created.
Specifically, with the three-dimensional reintegration method,
approximation is not necessary when degenerating three dimension to
two dimension or one dimension, thereby enabling high-precision
processing. Also, movement in an oblique direction may be processed
without degenerating to two dimension. Further, no degenerating to
two dimension enables process at each dimension. For example, with
the two-dimensional reintegration method, in the event of
degenerating in the spatial directions (X direction and Y
direction), process in the t direction serving as the temporal
direction cannot be performed. On the other hand, with the
three-dimensional reintegration method, any process in the
time-space directions may be performed.
Note that as described above, creation of any one of temporal
resolution and spatial resolution may be performed even with the
one-dimensional reintegration method, but creation of both temporal
resolution and spatial resolution cannot be performed with the
one-dimensional reintegration method in theory, which becomes
possible only by performing two-dimensional or more reintegration.
That is to say, creation of both temporal resolution and spatial
resolution becomes possible only by employing the above
two-dimensional reintegration method and the three-dimensional
reintegration method.
Also, the three-dimensional reintegration method takes not
one-dimensional and two-dimensional but three-dimensional
integration effects into consideration, and accordingly, an image
more similar to the light signal in the actual world 1 (FIG. 259)
may be created.
In other words, with the three-dimensional reintegration method,
for example, the actual world estimating unit 102 in FIG. 259 (FIG.
3) estimates the light signal function F representing the light
signals in the actual world by approximating the light signal
function F with a predetermined approximation function f on
condition that, the pixel value of a pixel corresponding to at
least a position in the one-dimensional direction of the time-space
directions, of an input image made up of multiple pixels having a
pixel value on which the light signals in the actual world 1 are
projected by the multiple detecting elements of the sensor 2 each
having spatio-temporal integration effects, and projected by the
detecting elements of which a part of continuity of the light
signals in the actual world 1 drops, is a pixel value acquired by
at least integration effects in the one-dimensional direction,
which is an assumption.
Further, for example, in the event that the data continuity
detecting unit 101 in FIG. 259 (FIG. 3) detects continuity of data
of an input image, the actual world estimating unit 102 estimates
the light signal function F by approximating the light signal
function F with the approximation function f on condition that the
pixel value of a pixel corresponding to at least a position in the
one-dimensional direction in the time-space directions of the image
data, corresponding to continuity of data detected by the data
continuity detecting unit 101 is the pixel value acquired by at
least integration effects in the one-dimensional direction, which
is an assumption.
Speaking in detail, for example, the actual world estimating unit
102 estimates the light signal function by approximating the light
signal function F with an approximation function on condition that
the pixel value of a pixel corresponding to at least a distance
along in the one-dimensional direction from a line corresponding to
continuity of data detected by the continuity detecting nit 101 is
the pixel value acquired by at least integration effects in the
one-dimensional direction, which is an assumption.
With the three-dimensional reintegration method, for example, the
image generating unit 103 (configuration is FIG. 279) in FIG. 259
(FIG. 3) generates a pixel value corresponding to a pixel having a
desired size by integrating the light signal function F estimated
by the actual world estimating unit 102, i.e., the approximation
function f in at least desired increments in the one-dimensional
direction (e.g., by calculating the right side of the above
Expression (201)).
Accordingly, with the three-dimensional reintegration method, an
image more similar to the light signal in the actual world 1 (FIG.
259) than that in conventional image generating methods, or the
above one-dimensional or two-dimensional reintegration method may
be generated.
Next, description will be made regarding the image generating unit
103 which newly generates pixels based on the derivative value or
gradient of each pixel in the event that the actual world
estimating information input from the actual world estimating unit
102 is information of the derivative value or gradient of each
pixel on the approximation function f(x) approximately representing
each pixel value of reference pixels with reference to FIG.
281.
Note that the term "derivative value" mentioned here, following the
approximation function f(x) approximately representing each pixel
value of reference pixels being obtained, means a value obtained at
a predetermined position using a one-dimensional differential
equation f (x)' obtained from the approximation function f(x)
thereof (one-dimensional differential equation f (t)' obtained from
an approximation function f(t) in the event that the approximation
function is in the frame direction). Also, the term "gradient"
mentioned here means the gradient of a predetermined position on
the approximation function f(x) directly obtained from the pixel
values of perimeter pixels at the predetermined position without
obtaining the above approximation function f(x) (or f (t)).
However, derivative values mean the gradient at a predetermined
position on the approximation function f(x), and accordingly,
either case means the gradient at a predetermined position on the
approximation function f(x). Accordingly, with regard to derivative
values and a gradient serving as the actual world estimating
information input from the actual world estimating unit 102, they
are unified and referred to as the gradient on the approximation
function f(x) (or f (t)), with description of the image generating
unit 103 in FIG. 281 and FIG. 285.
A gradient acquiring unit 3401 acquires the gradient information of
each pixel, the pixel value of the corresponding pixel, and the
gradient in the direction of continuity regarding the approximation
function f(x) approximately representing the pixel values of the
reference pixels input from the actual world estimating unit 102,
and outputs these to an extrapolation/interpolation unit 3402.
The extrapolation/interpolation unit 3402 generates certain-powered
higher-density pixels than an input image using
extrapolation/interpolation based on the gradient of each pixel on
the approximation function f(x), the pixel value of the
corresponding pixel, and the gradient in the direction of
continuity, which are input from the gradient acquiring unit 3401,
and outputs the pixels as an output image.
Next, description will be made regarding image generating
processing by the image generating unit 103 in FIG. 281 with
reference to the flowchart in FIG. 282.
In step S3401, the gradient acquiring unit 3401 acquires
information regarding the gradient (derivative value) on the
approximation function f(x), position, and pixel value of each
pixel, and the gradient in the direction of continuity, which is
input from the actual world estimating unit 102, as actual world
estimating information.
At this time, for example, in the event of generating an image made
up of pixels having double density in the spatial direction X and
spatial direction Y (quadruple in total) as to an input image,
information regarding as to a pixel Pin such as shown in FIG. 283,
gradients f (Xin)' (gradient in the center position of the pixel
Pin), f (Xin-Cx (-0.25))' (gradient of the center position of a
pixel Pa when generating a pixel of double density in the Y
direction from the pixel Pin), and f (Xin-Cx (0.25))' (gradient of
the center position of a pixel Pb when generating a pixel of double
density in the Y direction from the pixel Pin), the position and
pixel value of the pixel Pin, and a gradient G.sub.f in the
direction of continuity is input from the actual world estimating
unit 102.
In step S3402, the gradient acquiring unit 3401 selects information
of the corresponding pixel of interest, of the actual world
estimating information input, and outputs this to the
extrapolation/interpolation unit 3402.
In step S3403, the extrapolation/interpolation unit 3402 obtains a
shift amount from the position information of the input pixels, and
the gradient G.sub.f in the direction of continuity.
Here, a shift amount Cx (ty) is defined as Cx (ty)=ty/G.sub.f when
the gradient as continuity is represented with G.sub.f. This shift
amount Cx (ty) represents a shift width as to the spatial direction
X at a position in the spatial direction Y=ty of the approximation
function f(x), which is defined on the position in the spatial
direction Y=0. Accordingly, for example, in the event that an
approximation function on the position in the spatial direction Y=0
is defined as f (x), in the spatial direction Y=ty this
approximation function f(x) becomes a function shifted by the Cx
(ty) as to the spatial direction X, so that this approximation
function is defined as f (x-Cx(ty)) (=f(x-ty/G.sub.f).
For example, in the event of the pixel Pin such as shown in FIG.
283, when one pixel (one pixel size in the drawing is 1 both in the
horizontal direction and in the vertical direction) in the drawing
is divided into two pixels in the vertical direction (when
generating a double-density pixel in the vertical direction), the
extrapolation/interpolation unit 3402 obtains the shift amounts of
the pixels Pa and Pb, which are to be obtained. That is to say, in
this case, the pixels Pa and Pb are shifted by -0.25 and 0.25 as to
the spatial direction Y respectively as viewed from the pixel Pin,
so that the shift amounts of the pixels Pa and Pb become Cx (-0.25)
and Cx (0.25) respectively. Note that in FIG. 283, the pixel Pin is
a square of which general gravity position is (Xin, Yin), and the
pixels Pa and Pb are rectangles long in the horizontal direction in
the drawing of which general gravity positions are (Xin, Yin+0.25)
and (Xin, Yin-0.25) respectively.
In step S3404, the extrapolation/interpolation unit 3402 obtains
the pixel values of the pixels Pa and Pb using
extrapolation/interpolation through the following Expression (206)
and Expression (207) based on the shift amount Cx obtained at the
processing in step S3403, the gradient f (Xin)' on the pixel of
interest on the approximation function f(x) of the pixel Pin
acquired as the actual world estimating information, and the pixel
value of the pixel Pin. Pa=Pin-f(Xin)'.times.Cx(0.25) (206)
Pb=Pin-f(Xin)'.times.Cx(-0.25) (207)
In the above Expression (206) and Expression (207), Pa, Pb, and Pin
represent the pixel values of the pixels Pa, Pb, and Pin
respectively.
That is to say, as shown in FIG. 284, the amount of change of the
pixel value is set by multiplying the gradient f (Xin)' in the
pixel of interest Pin by the movement distance in the X direction,
i.e., shift amount, and the pixel value of a pixel to be newly
generated is set on the basis of the pixel value of the pixel of
interest.
In step S3405, the extrapolation/interpolation unit 3402 determines
regarding whether or not pixels having predetermined resolution
have been obtained. For example, in the event that predetermined
resolution is pixels having double density in the vertical
direction as to the pixels in an input image, the
extrapolation/interpolation unit 3402 determines that pixels having
predetermined resolution have been obtained by the above
processing, but for example, in the event that pixels having
quadruple density (double in the horizontal direction.times.double
in the vertical direction) as to the pixels in the input image have
been desired, pixels having predetermined resolution have not been
obtained by the above processing. Consequently, in the event that a
quadruple-density image is a desired image, the
extrapolation/interpolation unit 3402 determines that pixels having
predetermined resolution have not been obtained, and the processing
returns to step S3403.
In step S3403, the extrapolation/interpolation unit 3402 obtains
the shift amounts of pixels P01, P02, P03, and P04 (pixel having
quadruple density as to the pixel of interest Pin), which are to be
obtained, from the center position of a pixel, which is to be
generated, at the second processing respectively. That is to say,
in this case, the pixels P01 and P02 are pixels to be obtained from
the pixel Pa, so that each shift amount from the pixel Pa is
obtained respectively. Here, the pixels P01 and P02 are shifted by
-0.25 and 0.25 as to the spatial direction X respectively as viewed
from the pixel Pa, and accordingly, each value itself becomes the
shift amount thereof (since the pixels are shifted as to the
spatial direction X). Similarly, the pixels P03 and P04 are shifted
by -0.25 and 0.25 respectively as to the spatial direction X as
viewed from the pixel Pb, and accordingly, each value itself
becomes the shift amount thereof. Note that in FIG. 283, the pixels
P01, P02, P03, and P04 are squares of which gravity positions are
four cross-marked positions in the drawing, and the length of each
side is 1 for the pixel Pin, and accordingly, around 5 for the
pixels P01, P02, P03, and P04 respectively.
In step S3404, the extrapolation/interpolation unit 3402 obtains
the pixel values of the pixels P01, P02, P03, and P04 using
extrapolation/interpolation through the following Expression (208)
through Expression (211) based on the shift amount Cx obtained at
the processing in step S3403, the gradients f (Xin-Cx(-0.25))' and
f (Xin-Cx(0.25))' at a predetermined position on the approximation
function f(x) of the pixels Pa and Pb acquired as actual world
estimating information, and the pixel values of the pixels Pa and
Pb obtained at the above processing, and stores these in unshown
memory. P01=Pa+f(Xin-Cx(0.25))'.times.(-0.25) (208)
P02=Pa+f(Xin-Cx(0.25))'.times.(0.25) (209)
P03=Pb+f(Xin-Cx(-0.25))'.times.(-0.25) (210)
P04=Pb+f(Xin-Cx(-0.25))'.times.(0.25) (211)
In the above Expression (208) through Expression (211), P01 through
P04 represent the pixel values of the pixels P01 through P04
respectively.
In step S3405, the extrapolation/interpolation unit 3402 determines
regarding whether or not pixels having predetermined resolution
have been obtained, and in this case, the desired quadruple-density
pixels have been obtained, and accordingly, the
extrapolation/interpolation unit 3402 determines that the pixels
having predetermined resolution have been obtained, and the
processing proceeds to step S3406.
In step S3406, the gradient acquiring unit 3401 determines
regarding whether or not the processing of all pixels has been
completed, and in the event that determination is made that the
processing of all pixels has not been completed, the processing
returns to step S3402, wherein the subsequent processing is
repeatedly performed.
In step S3406, in the event that the gradient acquiring unit 3401
determines that the processing of all pixels has been completed,
the extrapolation/interpolation unit 3402 outputs an image made up
of the generated pixels, which are stored in unshown memory, in
step S3407.
That is to say, as shown in FIG. 284, the pixel values of new
pixels are obtained using extrapolation/interpolation according to
a distance apart in the spatial direction X from the pixel of
interest of which gradient is obtained using the gradient f (x)' on
the approximation function f(x).
Note that with the above example, description has been made
regarding the gradient (derivative value) at the time of
calculating a quadruple-density pixel as an example, but in the
event that gradient information at many more positions can be
obtained as the actual world estimating information, pixels having
more density in the spatial directions than that in the above
example may be calculated using the same method as the above
example.
Also, with regard to the above example, description has been made
regarding an example for obtaining double-density pixel values, but
the approximation function f(x) is a continuous function, and
accordingly, in the event that necessary gradient (derivative
value) information can be obtained even regarding pixel values
having density other than double density, an image made up of
further high-density pixels may be generated.
According to the above description, based on the gradient (or
derivative value) f (x)' information of the approximation function
f(x) approximating the pixel value of each pixel of an input image
supplied as the actual world estimating information in the spatial
direction, the pixels of an higher resolution image than the input
image may be generated.
Next, description will be made with reference to FIG. 285 regarding
the image generating unit 103 for generating new pixel values so as
to output an image based upon the derivative values or gradient
information for each pixel in a case that the actual world
estimation information input from the actual world estimating unit
102 is derivative values or gradient information for these pixels,
obtained from f(t) that is a function in the frame direction (time
direction) representing approximate pixel values of the reference
pixels.
An gradient acquisition unit 3411 acquires the gradient information
obtained from an approximate function f(t) which represents
approximate pixel values of the reference pixels, the corresponding
pixel value, and movement as continuity, for each pixel position,
which are input from the actual world estimating unit 102, and
outputs the information thus obtained to an extrapolation unit
3412.
The extrapolation unit 3412 generates a high-density pixel of a
predetermined order higher than that of the input image using
extrapolation based upon the gradient which is obtained from the
approximate function f(t), the corresponding pixel value, and
movement as continuity, for each pixel, which are input from the
gradient acquisition unit 3411, and outputs the image thus
generated as an output image.
Next, description will be made regarding image generating
processing by the image generating unit 103 shown in FIG. 285, with
reference to the flowchart shown in FIG. 286.
In Step S3421, the gradient acquisition unit 3411 acquires
information regarding the gradient (derivative value) which is
obtained from the approximate function f(t), the position, the
pixel value, and movement as continuity, for each pixel, which are
input from the actual world estimating unit 102, as actual world
estimation information.
For example, in a case of generating an image from the input image
with double pixel density in both the spatial direction and the
frame direction (i.e., a total of quadruple pixel density), the
input information regarding the pixel Pin shown in FIG. 287,
received from the actual world estimating unit 102 includes: the
gradient f(Tin)' (the gradient at the center of the pixel Pin),
f(Tin-Ct(0.25))' (the gradient at the center of the pixel Pat
generated in a step for generating pixels in the Y direction from
the pixel Pin with double pixel density), f(Tin-Ct(-0.25))' (the
gradient at the center of the pixel Pbt generated in a step for
generating pixels in the Y direction from the pixel Pin with double
pixel density), the position of the pixel Pin, the pixel value, and
movement as continuity (motion vector).
In Step S3422, the gradient acquisition unit 3411 selects the
information regarding the pixel of interest, from the input actual
world estimation information, and outputs the information thus
acquired, to the extrapolation unit 3412.
In Step S3423, the extrapolation unit 3412 calculates the shift
amount based upon the position information thus input, regarding
the pixel and the gradient of continuity direction.
Here, with movement as continuity (gradient on the plane having the
frame direction and the spatial direction) as V.sub.f, the shift
amount Ct(ty) is obtained by the equation Ct(ty)=ty/V.sub.f. The
shift amount Ct(ty) represents the shift of the approximate
function f(t) in the frame direction T, calculated at the position
of Y=ty in the spatial direction. Note that the approximate
function f(t) is defined at the position Y=0 in the spatial
direction. Accordingly, in a case that the approximate function
f(t) is defined at the position Y=0 in the spatial direction, for
example, the approximate function f(t) is shifted at Y=ty in the
spatial direction by Ct(ty) in the spatial direction T, and
accordingly, the approximate function at Y=ty is defined as
f(t-Ct(ty)) (=f(t-ty/V.sub.f)).
For example, let us consider the pixel Pin as shown in FIG. 287. In
a case that the one pixel in the drawing (let us say that the pixel
is formed with a pixel size of (1, 1) both in the frame direction
and the spatial direction) is divided into two in the spatial
direction (in a case of generating an image with double pixel
density in the spatial direction), the extrapolation unit 3412
calculates the shift amounts for obtaining the pixels Pat and Pbt.
That is to say, the pixels Pat and Pbt are shifted along the
spatial direction Y from the pixel Pin by 0.25 and -0.25,
respectively. Accordingly, the shift amounts for obtaining the
pixel values of the pixels Pat and Pbt are Ct(-0.25) and Ct(0.25),
respectively. Note that in FIG. 287, the pixel Pin is formed in the
shape of a square with the center of gravity at around (Xin, Yin).
On the other hand, the pixels Pat and Pbt are formed in the shape
of a rectangle having long sides in the horizontal direction in the
drawing with the centers of gravity of around (Xin, Yin+0.25) and
(Xin, Yin-0.25), respectively.
In Step S3424, the extrapolation unit 3412 calculates the pixel
values of the pixels Pat and Pbt with the following Expressions
(212) and (213) using extrapolation based upon the shift amount
obtained in Step S3423, the gradient f(Tin)' at the pixel of
interest, which is obtained from the approximate function f(t) for
providing the pixel value of the pixel Pin and has been acquired as
the actual world estimation information, and the pixel value of the
pixel Pin. pat=Pin-f(Tin)'.times.Ct(0.25) (212)
pbt=Pin-f(Xin)'.times.Ct(-0.25) (213)
In the above Expressions (212) and (213), Pat, Pbt, and Pin
represent the pixel values of the pixel Pat, Pbt, and Pin,
respectively.
That is to say, as shown in FIG. 288, the change in the pixel value
is calculated by multiplying the gradient f(Xin)' at the pixel of
interest Pin by the distance in the X direction, i.e., the shift
amount. Then, the value of a new pixel, which is to be generated,
is determined using the change thus calculated with the pixel value
of the pixel of interest as a base.
In Step S3425, the extrapolation unit 3412 determines whether or
not the pixels thus generated provide requested resolution. For
example, in a case that the user has requested resolution of double
pixel density in the spatial direction as compared with the input
image, the extrapolation unit 3412 determines that requested
resolution image has been obtained. However, in a case that the
user has requested resolution of quadruple pixel density (double
pixel density in both the frame direction and the spatial
direction), the above processing does not provide the requested
pixel density. Accordingly, in a case that the user has requested
resolution of quadruple pixel density, the extrapolation unit 3412
determines that requested resolution image has not been obtained,
and the flow returns to Step S3423.
In Step S3423 for the second processing, the extrapolation unit
3412 calculates the shift amounts from the pixels as bases for
obtaining the centers of the pixels P01t, P02t, P03t, and P04t
(quadruple pixel density as compared with the pixel of interest
Pin). That is to say, in this case, the pixels P01t and P02t are
obtained from the pixel Pat, and accordingly, the shift amounts
from the pixel Pat are calculated for obtaining these pixels. Here,
the pixels P01t and P02t are shifted from the pixel Pat in the
frame direction T by -0.25 and 0.25, respectively, and accordingly,
the distances therebetween without any conversion are employed as
the shift amounts. In the same way, the pixels P03t and P04t are
shifted from the pixel Pbt in the frame direction T by -0.25 and
0.25, respectively, and accordingly, the distances therebetween
without any conversion are employed as the shift amounts. Note that
in FIG. 287, each of the pixels P01t, P02t, P03t, and P04t is
formed in the shape of a square having the center of gravity
denoted by a corresponding one of the four cross marks in the
drawing, and the length of each side of each of these pixels P01t,
P02t, P03t, and P04t is approximately 0.5, since the length of each
side of the pixel Pin is 1.
In Step S3424, the extrapolation unit 3412 calculates the pixel
values of the pixels P01t, P02t, P03t, and P04t, with the following
Expressions (214) through (217) using extrapolation based upon the
shift amount Ct obtained in Step S3423, f(Tin-Ct(0.25))' and
f(Tin-Ct(-0.25))' which are the gradients of the approximate
function f(t) at the corresponding positions of Pat and Pbt and
acquired as the actual world estimation information, and the pixel
values of the pixels Pat and Pbt obtained in the above processing.
The pixel values of the pixels P01t, P02t, P03t, and P04t thus
obtained are stored in unshown memory.
P01t=Pat+f(Tin-Ct(0.25))'.times.(-0.25) (214)
P02t=Pat+f(Tin-Ct(0.25))'.times.(0.25) (215)
P03t=Pbt+f(Tin-Ct(-0.25))'.times.(-0.25) (216)
P04t=Pbt+f(Tin-Ct(-0.25))'.times.(0.25) (217)
In the above Expressions (208) through (211), P01t through P04t
represent the pixel values of the pixels P01t through P04t,
respectively.
In Step S3425, the extrapolation unit 3412 determines whether or
not the pixel density for achieving the requested resolution has
been obtained. In this stage, the requested quadruple pixel density
is obtained. Accordingly, the extrapolation unit 3412 determines
that the pixel density for requested resolution has been obtained,
following which the flow proceeds to Step S3426.
In Step S3426, the gradient acquisition unit 3411 determines
whether or not processing has been performed for all the pixels. In
a case that the gradient acquisition unit 3411 determines that
processing has not been performed for all the pixels, the flow
returns to Step S3422, and subsequent processing is repeated.
In Step S3426, the gradient acquisition unit 3411 determines that
processing has been performed for all the pixels, the extrapolation
unit 3412 outputs an image formed of generated pixels stored in the
unshown memory in Step S3427.
That is to say, as shown in FIG. 288, the gradient of the pixel of
interest is obtained using the gradient f(t)' of the approximate
function f(t), and the pixel values of new pixels are calculated
corresponding to the number of frames positioned along the frame
direction T from the pixel of interest.
While description has been made in the above example regarding an
example of the gradient (derivative value) at the time of computing
a quadruple-density pixel, the same technique can be used to
further compute pixels in the frame direction as well, if gradient
information at a greater number of positions can be obtained as
actual world estimation information.
While description has been made regarding an arrangement for
obtaining a double pixel-density image, an arrangement may be made
wherein much higher pixel-density image is obtained based upon the
information regarding the necessary gradient information
(derivative values) using the nature of the approximate function
f(t) as a continuous function.
The above-described processing enables creation of a higher
resolution pixel image than the input image in the frame direction
based upon the information regarding f(t)' which is supplied as the
actual world estimation information, and is the gradient (or
derivative value) of the approximate function f(t) which provides
an approximate value of the pixel value of each pixel of the input
image.
With the present embodiment described above, data continuity is
detected from the image data formed of multiple pixels having the
pixel values obtained by projecting the optical signals in the real
world by actions of multiple detecting elements; a part of
continuity of the optical signals in the real world being lost due
to the projection with the multiple detecting elements each of
which has time-space integration effects. Then, the gradients at
the multiple pixels shifted from the pixel of interest in the image
data in one dimensional direction of the time-space directions are
employed as a function corresponding to the optical signals in the
real world. Subsequently, the line is calculated for each of the
aforementioned multiple pixels shifted from the center of the pixel
of interest in the predetermined direction, with the center
matching that of the corresponding pixel and with the gradient at
the pixel thus employed. Then, the values at both ends of the line
thus obtained within the pixel of interest are employed as the
pixel values of a higher pixel-density image than the input image
formed of the pixel of interest. This enables creation of
high-resolution image in the time-space directions than the input
image.
Next, description will be made regarding another arrangement of the
image generating unit 103 (see FIG. 3) according to the present
embodiment with reference to FIG. 289 through FIG. 314.
FIG. 289 shows an example of a configuration of the image
generating unit 103 according to the present embodiment.
The image generating unit 103 shown in FIG. 289 includes a class
classification adaptation unit 3501 for executing conventional
class classification adaptation processing, a class classification
adaptation correction unit 3502 for performing correction of the
results of the class classification adaptation processing (detailed
description will be made later), and addition unit 3503 for making
the sum of an image output from the class classification adaptation
unit 3501 and an image output from the class classification
adaptation processing correction unit 3502, and outputting the
summed image as an output image to external circuits.
Note that the image output from the class classification adaptation
processing unit 3501 will be referred to as "predicted image"
hereafter. On the other hand, the image output from the class
classification adaptation processing correction unit 3502 will be
referred to as "correction image" or "subtraction predicted image".
Note that description will be made later regarding the concept
behind the "predicted image" and "subtraction predicted image".
Also, in the present embodiment, let us say that the class
classification adaptation processing is processing for improving
the spatial resolution of the input image, for example. That is to
say, the class classification adaptation processing is processing
for converting the input image with standard resolution into the
predicted image with high resolution.
Note that the image with the standard resolution will be referred
to as "SD (Standard Definition) image" hereafter as appropriate.
Also, the pixels forming the SD image will be referred to as "SD
pixels" as appropriate.
On the other hand, the high-resolution image will be referred to as
"HD (High Definition) image" hereafter as appropriate. Also, the
pixels forming the HD image will be referred to as "HD pixels" as
appropriate.
Next, description will be made below regarding a specific example
of the class classification adaptation processing according to the
present embodiment.
First, the features are obtained for each of the SD pixels
including the pixel of interest and the pixels therearound (such SD
pixels will be referred to as "class tap" hereafter) for
calculating the HD pixels of the predicted image (HD image)
corresponding to the pixel of interest (SD pixel) of the input
image (SD image). Then, the class of the class tap is selected from
classes prepared beforehand, based upon the features thus obtained
(the class code of the class tap is determined).
Then, product-sum calculation is performed using the coefficients
forming a coefficient set selected from multiple coefficient sets
prepared beforehand (each coefficient set corresponds to a certain
class code) based upon the class code thus determined, and the SD
pixels including the pixel of interest and the pixels therearound
(Such SD pixels will be referred to as "prediction tap" hereafter.
Note that the class tap may also be employed as the prediction
tap.), so as to obtain HD pixels of a predicted image (HD image)
corresponding to the pixel of interest (SD pixel) of the input
image (SD image).
Accordingly, with the arrangement according to the present
embodiment, the input image (SD image) is subjected to conventional
class classification adaptation processing at the class
classification adaptation processing unit 3501 so as to generate
the predicted image (HD image). Furthermore, the predicted image
thus obtained is corrected at the addition unit 3503 using the
correction image output from the class classification adaptation
processing correction unit 3502 (by making the sum of the predicted
image and the correction image), thereby obtaining the output image
(HD image).
That is to say, the arrangement according to the present embodiment
can be said to be an arrangement of the image generating unit 103
of the image processing device (FIG. 3) for performing processing
based upon the continuity, from the perspective of the continuity.
On the other hand, the arrangement according to the present
embodiment can also be said to be an arrangement of the image
processing device further including the data continuity detecting
unit 101, the actual world estimating unit 102, the class
classification adaptation correction unit 3502, and the addition
unit 3503, for performing correction of the class classification
adaptation processing, as compared with a conventional image
processing device formed of the sensor 2 and the class
classification adaptation processing unit 3501, from the
perspective of class classification adaptation processing.
Accordingly, such an arrangement according to the present
embodiment will be referred to as "class classification processing
correction means" hereafter, as opposed to reintegration means
described above.
Detailed description will be made regarding the image generating
unit 103 using the class classification processing correction
means.
In FIG. 289, upon input of signals in the actual world 1
(distribution of the light intensity) to the sensor 2, the input
image is output from the sensor 2. The input image is input to the
class classification adaptation processing unit 3501 of the image
generating unit 103, as well as to the data continuity detecting
unit 101.
The class classification adaptation processing unit 3501 performs
conventional class classification adaptation processing for the
input image so as to generate the predicted image, and output the
predicted image to the addition unit 3503.
As described above, with the class classification adaptation
processing unit 3501, the input image (image data) input from the
sensor 2 is employed as a target image which is to be subjected to
processing, as well as a reference image. That is to say, although
the input image from the sensor 2 is different (distorted) from the
signals of the actual world 1 due to the integration effects
described above, the class classification adaptation processing
unit 3501 performs the processing using the input image different
from the signals of the actual world 1, as a correct reference
image.
As a result, in a case that the HD image is generated using the
class classification adaptation processing based upon the input
image (SD image) in which original details have been lost in the
input stage where the input image has been output from the sensor
2, such an HD image may have a problem that original details cannot
be reproduced completely.
In order to solve the aforementioned problem, with the class
classification processing correction means, the class
classification adaptation processing correction unit 3502 of the
image generating unit 103 employs the information (actual world
estimation information) for estimating the original image (signals
of the actual world 1 having original continuity) which is to be
input to the sensor 2, as a target image to be subjected to
processing as well as a reference image, instead of the input image
from the sensor 2, so as to create a correction image for
correcting the predicted image output from the class classification
adaptation processing unit 3501.
The actual world estimation information is created by actions of
the data continuity detecting unit 101 and the actual world
estimating unit 102.
That is to say, the data continuity detecting unit 101 detects the
continuity of the data (the data continuity corresponding to the
continuity contained in signals of the actual world 1, which are
input to the sensor 2) contained in the input image output from the
sensor 2, and outputs the detection results as the data continuity
information, to the actual world estimating unit 102.
Note that while FIG. 289 shows an arrangement wherein the angle is
employed as the data continuity information, the data continuity
information is not restricted to the angle, rather various kinds
information may be employed as the data continuity information.
The actual world estimating unit 102 creates the actual estimation
information based upon the angle (data continuity information) thus
input, and outputs the actual world estimation information thus
created, to the class classification adaptation correction unit
3502 of the image generating unit 103.
Note that while FIG. 289 shows an arrangement wherein the
features-amount image (detailed description thereof will be made
later) is employed as the actual world estimation information, the
actual world estimation information is not restricted to the
features-amount image, various information may be employed as
described above.
The class classification adaptation processing correction unit 3502
creates a correction image based upon the features-amount image
(actual world estimation information) thus input, and outputs the
correction image to the addition unit 3503.
The addition unit 3503 makes the sum of the predicted image output
from the class classification adaptation processing unit 3501 and
the correction image output from the class classification
adaptation processing correction unit 3502, and outputs the summed
image (HD image) as an output image, to external circuits.
The output image thus output is similar to the signals (image) of
the actual world 1 with higher precision than the predicted image.
That is to say, the class classification adaptation processing
correction means enable the user to solve the aforementioned
problem.
Furthermore, with the signal processing device (image processing
device) 4 having a configuration as shown in FIG. 289, such
processing can be applied for the entire area of one frame. That is
to say, while a signal processing device using a hybrid technique
described later (e.g., an arrangement described later with
reference to FIG. 315) or the like has need of identifying the
pixel region for generating the output image, the signal processing
device 4 shown in FIG. 289 has the advantage that there is no need
of identifying such pixel region.
Next, description will be made in detail regarding the class
classification adaptation processing unit 3510 of the image
generating device 103.
FIG. 290 shows a configuration example of the class classification
adaptation processing unit 3501.
In FIG. 290, the input image (SD image) input from the sensor 2 is
supplied to a region extracting unit 3511 and a region extracting
unit 3515. The region extracting unit 3511 extracts a class tap
(the SD pixels existing at predetermined positions, which includes
the pixel of interest (SD pixel)), and outputs the class tap to a
pattern detecting unit 3512. The pattern detecting unit 3512
detects the pattern of the input image based upon the class tap
thus input.
A class-code determining unit 3513 determines the class code based
upon the pattern detected by the pattern detecting unit 3512, and
outputs the class code to a coefficient memory 3514 and a region
extracting unit 3515. The coefficient memory 3514 stores the
coefficients for each class code prepared beforehand by learning,
reads out the coefficients corresponding to the class code input
from the class code determining unit 3513, and outputs the
coefficients to a prediction computing unit 3516.
Note that description will be made later regarding the learning
processing for obtaining the coefficients stored in the coefficient
memory 3514, with reference to a block diagram of a class
classification adaptation processing learning unit shown in FIG.
292.
Also, the coefficients stored in the coefficient memory 3514 are
used for creating a prediction image (HD image) as described later.
Accordingly, the coefficients stored in the coefficient memory 3514
will be referred to as "prediction coefficients" in order to
distinguishing the aforementioned coefficients from other kinds of
coefficients.
The region extracting unit 3515 extracts a prediction tap (SD
pixels which exist at predetermined positions including the pixel
of interest) necessary for predicting and creating a prediction
image (HD image) from the input image (SD image) input from the
sensor 2 based upon the class code input from the class code
determining unit 3513, and outputs the prediction tap to the
prediction computing unit 3516.
The prediction computing unit 3516 executes product-sum computation
using the prediction tap input from the region extracting unit 3515
and the prediction coefficients input from the coefficient memory
3514, creates the HD pixels of the prediction image (HD image)
corresponding to the pixel of interest (SD pixel) of the input
image (SD image), and outputs the HD pixels to the addition unit
3503.
More specifically, the coefficient memory 3514 outputs the
prediction coefficients corresponding to the class code supplied
from the class code determining unit 3513 to the prediction
computing unit 3516. The prediction computing unit 3516 executes
the product-sum computation represented by the following Expression
(218) using the prediction tap which is supplied from the region
extracting unit 3515 and is extracted from the pixel values of
predetermined pixels of the input image, and the prediction
coefficients supplied from the coefficient memory 3514, thereby
obtaining (predicting and estimating) the HD pixels of the
prediction image (HD image).
'.times..times. ##EQU00132##
In Expression (218), q' represents the HD pixel of the prediction
image (HD image). Each of c.sub.i (i represents an integer of 1
through n) represents the corresponding prediction tap (SD pixel).
Furthermore, each of d.sub.i represents the corresponding
prediction coefficient.
As described above, the class classification adaptation processing
unit 3501 predicts and estimates the corresponding HD image based
upon the SD image (input image), and accordingly, in this case, the
HD image output from the class classification adaptation processing
unit 3501 is referred to as "prediction image".
FIG. 291 shows a learning device (calculating device for obtaining
the prediction coefficients) for determining the prediction
coefficients (d.sub.i in Expression (215)) stored in the
coefficient memory 3514 of the class classification adaptation
processing unit 3501.
Note that with the class classification adaptation processing
correction technique, coefficient memory (correction coefficient
memory 3554 which will be described later with reference to FIG.
299) is included in the class classification adaptation processing
correction unit 3502, in addition to the coefficient memory 3514.
Accordingly, as shown in FIG. 291, a learning device 3504 according
to the class classification adaptation processing technique
includes a learning unit 3561 (which will be referred to as "class
classification adaptation processing correction learning unit 3561"
hereafter) for determining the coefficients stored in the
correction coefficient memory 3554 of the class classification
adaptation processing correction unit 3502 as well as a learning
unit 3521 (which will be referred to as "class classification
adaptation processing learning unit 3521" hereafter) for
determining the prediction coefficients (d.sub.i in Expression
(215)) stored in the coefficient memory 3514 of the class
classification adaptation processing unit 3501.
Accordingly, while the tutor image used in the class classification
adaptation processing learning unit 3521 will be referred to as
"first tutor image" hereafter, the tutor image used in the class
classification adaptation processing correction learning unit 3561
will be referred to as "second tutor image" hereafter. In the same
way, while the student image used in the class classification
adaptation processing learning unit 3521 will be referred to as
"first student image" hereafter, the student image used in the
class classification adaptation processing correction learning unit
3561 will be referred to as "second student image" hereafter.
Note that description will be made later regarding the class
classification adaptation processing correction learning unit
3561.
FIG. 292 shows a detailed configuration example of the class
classification adaptation processing learning unit 3521.
In FIG. 292, a certain image is input to the class classification
adaptation processing correction learning unit 3561 (FIG. 291), as
well as to a down-converter unit 3531 and a normal equation
generating unit 3536 as a first tutor image (HD image).
The down-converter unit 3531 generates a first student image (SD
image) with a lower resolution than the first tutor image based
upon the input first tutor image (HD image) (converts the first
tutor image into a first student image with a lower resolution.),
and outputs the first student image to region extracting units 3532
and 3535, and the class classification adaptation processing
correction learning unit 3561 (FIG. 291).
As described above, the class classification adaptation processing
learning unit 3521 includes the down-converter unit 3531, and
accordingly, the first tutor image (HD image) has no need of having
a higher resolution than the input image from the aforementioned
sensor 2 (FIG. 289). The reason is that in this case, the first
tutor image subjected to down-converting processing (the processing
for reducing the resolution of the image) is employed as the first
student image, i.e., the SD image. That is to say, the first tutor
image corresponding to the first student image is employed as an HD
image. Accordingly, the input image from the sensor 2 may be
employed as the first tutor image without any conversion.
The region extracting unit 3532 extracts the class tap (SD pixels)
necessary for class classification from the first student image (SD
image) thus supplied, and outputs the class tap to a pattern
detecting unit 3533. The pattern detecting unit 3533 detects the
pattern of the class tap thus input, and outputs the detection
results to a class code determining unit 3534. The class code
determining unit 3534 determines the class code corresponding to
the input pattern, and outputs the class code to the region
extracting unit 3535 and the normal equation generating unit
3536.
The region extracting unit 3535 extracts the prediction tap (SD
pixels) from the first student image (SD image) input from the
down-converter unit 3531 based upon the class code input from the
class code determining unit 3534, and outputs the prediction tap to
the normal equation generating unit 3536 and a prediction computing
unit 3558.
Note that the region extracting unit 3532, the pattern detecting
unit 3533, the class-code determining unit 3534, and the region
extracting unit 3535 have generally the same configurations and
functions as those of the region extracting unit 3511, the pattern
detecting unit 3512, the class-code determining unit 3513, and the
region extracting unit 3515, of the class classification adaptation
processing unit 3501 shown in FIG. 290.
The normal equation generating unit 3536 generates normal equations
based upon the prediction tap (SD pixels) of the first student
image (SD image) input from the region extracting unit 3535, and
the HD pixels of the first tutor image (HD image), for each class
code of all class codes input form the class code determining unit
3545, and supplies the normal equations to a coefficient
determining unit 3537. Upon reception of the normal equations
corresponding to a certain class code from the normal equation
generating unit 3537, the coefficient determining unit 3537
computes the prediction coefficients using the normal equations.
Then, the coefficient determining unit 3537 supplies the computed
prediction coefficients to a prediction computing unit 3538, as
well as storing the prediction coefficients in the coefficient
memory 3514 in association with the class code.
Detailed description will be made regarding the normal equation
generating unit 3536 and the coefficient determining unit 3537.
In the aforementioned Expression (218), each of the prediction
coefficients d.sub.i is undetermined coefficients before learning
processing. The learning processing is performed by inputting HD
pixels of the multiple tutor images (HD image) for each class code.
Let us say that there are m HD pixels corresponding to a certain
class code. With each of the m HD pixels as q.sub.k (k represents
an integer of 1 through m), the following Expression (219) is
introduced from the Expression (218).
.times..times..times..times. ##EQU00133##
That is to say, the Expression (219) indicates that the HD pixel
q.sub.k can be predicted and estimated by computing the right side
of the Expression (219). Note that in Expression (219), e.sub.k
represents error. That is to say, the HD pixel q.sub.k' which is a
prediction image (HD image) which is the results of computing the
right side, does not completely match the actual HD pixel q.sub.k,
and includes a certain error e.sub.k.
Accordingly, in Expression (219), the prediction coefficients
d.sub.i which exhibit the minimum of the sum of the squares of
errors e.sub.k should be obtained by the learning processing, for
example.
Specifically, the number of the HD pixels q.sub.k prepared for the
learning processing should be greater than n (i.e., m>n). In
this case, the prediction coefficients d.sub.i are determined as a
unique solution using the least square method.
That is to say, the normal equations for obtaining the prediction
coefficients d.sub.i in the right side of the Expression (219)
using the least square method are represented by the following
Expression (220).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times..times..times..times..-
times..times..times..times..times..times..times..times..times..times..time-
s..times..times.
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times. ##EQU00134##
Accordingly, the normal equations represented by the Expression
(220) are created and solved, thereby determining the prediction
coefficients d.sub.i as a unique solution.
Specifically, let us say that the matrices in the Expression (220)
representing the normal equations are defined as the following
Expressions (221) through (223). In this case, the normal equations
are represented by the following Expression (224).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times..times..times..times..-
times..times..times..times..times..times..times..times..times..times..time-
s..times..times..times..times..times..times..times..times..times..times..t-
imes..times..times..times..times..times..times..times..times..times.
##EQU00135##
As shown in Expression (222), each component of the matrix
D.sub.MAT is the prediction coefficient d.sub.i which is to be
obtained. With the present embodiment, the matrix C.sub.MAT in the
left side and the matrix Q.sub.MAT in the right side in Expression
(224) are determined, thereby obtaining the matrix D.sub.MAT (i.e.,
the prediction coefficients d.sub.i) using matrix computation.
More specifically, as shown in Expression (221), each component of
the matrix C.sub.MAT can be computed since the prediction tap
c.sub.ik is known. With the present embodiment, the prediction tap
c.sub.ik is extracted by the region extracting unit 3535. The
normal equation generating unit 3536 computes each component of the
matrix C.sub.MAT using the prediction tap c.sub.ik supplied from
the region extracting unit 3535.
Also, with the present embodiment, the prediction tap C.sub.ik and
the HD pixel q.sub.k are known. Accordingly, each component of the
matrix Q.sub.MAT can be computed as shown in Expression (223). Note
that the prediction tap C.sub.ik is the same as in the matrix
C.sub.MAT. Also, employed as the HD pixel q.sub.k is the HD pixel
of the first tutor image corresponding to the pixel of interest (SD
pixel of the first student image) included in the prediction tap
c.sub.ik. Accordingly, the normal equation generating unit 3536
computes each component of the matrix Q.sub.MAT based upon the
prediction tap c.sub.ik supplied from the region extracting unit
3535 and the first tutor image.
As described above, the normal equation generating unit 3536
computes each component of the matrix C.sub.MAT and the matrix
Q.sub.MAT, and supplies the computation results in association with
the class code to the coefficient determining unit 3537.
The coefficient determining unit 3537 computes the prediction
coefficient d.sub.i serving as each component of the matrix
D.sub.MAT in the above Expression (224) based upon the normal
equation corresponding to the supplied certain class code.
Specifically, the above Expression (224) can be transformed into
the following Expression (225)
.times. ##EQU00136##
In Expression (225), each component of the matrix D.sub.MAT in the
left side is the prediction coefficient d.sub.i which is to be
obtained. On the other hand, each component of the matrix C.sub.MAT
and the matrix Q.sub.MAT is supplied from the normal equation
generating unit 3536. With the present embodiment, upon reception
of each component of the matrix C.sub.MAT and the matrix Q.sub.MAT
corresponding to the current class code from the normal equation
generating unit 3536, the coefficient determining unit 3537
executes the matrix computation represented by the right side of
Expression (225), thereby computing the matrix D.sub.MAT. Then, the
coefficient determining unit 3537 supplies the computation results
(prediction coefficient d.sub.i) to the prediction computation unit
3538, as well as storing the computation results in the coefficient
memory 3514 in association with the class code.
The prediction computation unit 3538 executes product-sum
computation using the prediction tap input from the region
extracting unit 3535 and the prediction coefficients determined by
the coefficient determining unit 3537, thereby generating the HD
pixel of the prediction image (predicted image as the first tutor
image) corresponding to the pixel of interest (SD pixel) of the
first student image (SD image). The HD pixels thus generated are
output as a learning-prediction image to the class classification
adaptation processing correction learning unit 3561 (FIG. 291).
More specifically, with the prediction computation unit 3538, the
prediction tap extracted from the pixel values around a certain
pixel position in the first student image supplied from the region
extracting unit 3535 is employed as c.sub.i (i represents an
integer of 1 through n). Furthermore, each of the prediction
coefficients supplied from the coefficient determining unit 3537 is
employed as d.sub.i. The prediction computation unit 3538 executes
product-sum computation represented by the above Expression (218)
using the c.sub.i and d.sub.i thus employed, thereby obtaining the
HD pixel q' of the learning-prediction image (HD image) (i.e.,
thereby predicting and estimating the first tutor image).
Now, description will be made with reference to FIG. 293 through
FIG. 298 regarding a problem of the conventional class
classification adaptation processing (class classification
adaptation processing unit 3501) described above, i.e., a problem
that original details cannot be reproduced completely in a case
that the HD image (predicted image of signals in the actual world
1) is generated by the class classification adaptation processing
unit 3501 shown in FIG. 289 based upon the input image (SD image)
in which original details have been lost in the input stage where
the input image has been output from the sensor 2.
FIG. 293 shows an example of processing results of the class
classification adaptation unit 3501.
In FIG. 293, an HD image 3541 has a fine line with a gradient of
around 5.degree. clockwise as to the vertical direction in the
drawing. On the other hand, an SD image 3542 is generated from the
HD image 3541 such that the average of each block of 2.times.2
pixels (HD pixels) of the HD image 3541 is employed as the
corresponding single pixel (SD pixel) thereof. That is to say, the
SD image 3542 is "down-converted" (reduced-resolution) image of the
HD image 3541.
In other words, the HD image 3541 can be assumed to be an image
(signals in the actual world 1 (FIG. 289)) which is to be output
from the sensor 2 (FIG. 289) in this simulation. In this case, the
SD image 3542 can be assumed to be an image corresponding to the HD
image 3541, obtained from the sensor 2 having certain integration
properties in the spatial direction in this simulation. That is to
say, the SD image 3542 can be assumed to be an image input from the
sensor 2 in this simulation.
In this simulation, the SD image 3542 is input to the class
classification adaptation processing unit 3501 (FIG. 289). The
predicted image output from the class classification adaptation
processing unit 3501 is a predicted image 3543. That is to say, the
predicted image 3543 is an HD image (image with the same resolution
as with the original HD image 3541) generated by conventional class
classification adaptation processing. Note that the prediction
coefficients (prediction coefficients stored in the coefficient
memory 3514 (FIG. 290)) used for prediction computation by the
class classification adaptation processing unit 3501 are obtained
with learning/computation processing performed by the class
classification adaptation processing learning unit 3561 (FIG. 292)
with the HD image 3541 as the first tutor image and with the SD
image 3542 as the first student image.
Making a comparison between the HD image 3541, the SD image 3542,
and the predicted image 3543, it has been confirmed that the
predicted image 3543 is more similar to the HD image 3541 than the
SD image 3542.
The comparison results indicate that the class classification
adaptation processing 3501 generates the predicted image 3543 with
reproduced original details using conventional class classification
adaptation processing based upon the SD image 3542 in which the
original details in the HD image 3541 have been lost.
However, making a comparison between the predicted image 3543 and
the HD image 3541, it cannot be said definitely that the predicted
image 3543 is a complete reproduced image of the HD image 3541.
In order to investigate the cause of such insufficient reproduction
of the predicted image 3543 as to the HD image 3541, the present
applicant formed a summed image by making the sum of the HD image
3541 and the inverse image of the predicted image 3534 using the
addition unit 3546, i.e., a subtraction image 3544 obtained by
subtracting the predicted image 3543 from the HD image 3541 (In a
case of large difference in pixel values therebetween, the pixel of
the subtraction image is formed with a density close to white. On
the other hand, in a case of small difference in pixel values
therebetween, the pixel of the subtraction image is formed with a
density close to black.).
In the same way, the present applicant formed a summed image by
making the sum of the HD image 3541 and the inverse image of the SD
image 3542 using the addition unit 3547, i.e., a subtraction image
3545 obtained by subtracting the SD image 3542 from the HD image
3541 (In a case of large difference in pixel values therebetween,
the pixel of the subtraction image is formed with a density close
to white. On the other hand, in a case of small difference in pixel
values therebetween, the pixel of the subtraction image is formed
with a density close to black.).
Then, making a comparison between the subtraction image 3544 and
the subtraction image 3545, the present applicant obtained
investigation results as follows.
That is to say, the region which exhibits great difference in the
pixel value between the HD image 3541 and the SD image 3542 (i.e.,
the region formed with a density close to white, in the subtraction
image 3545) generally matches the region which exhibits great
difference in the pixel value between the HD image 3541 and the
predicted image 3543 (i.e., the region formed with a density close
to white, in the subtraction image 3544).
In other words, the region in the predicted image 3543, exhibiting
insufficient reproduction results as to the HD image 3541 generally
matches the region which exhibits great difference in the pixel
value between the HD image 3541 and the SD image 3542 (i.e., the
region formed with a density close to white, in the subtraction
image 3545).
Then, in order to solve the cause of the investigation results, the
present applicant further made investigation as follows.
That is to say, first, the present applicant investigated
reproduction results in the region which exhibits small difference
in the pixel value between the HD image 3541 and the predicted
image 3543 (i.e., the region formed with a density close to black,
in the subtraction image 3544). With the aforementioned region,
information obtained for this investigation are: the actual values
of the HD image 3541; the actual pixel values of the SD image 3542;
and the actual waveform corresponding to the HD image 3541 (signals
in the actual world 1). The investigation results are shown in FIG.
294 and FIG. 295.
FIG. 294 shows an example of the investigation-target region. Note
that in FIG. 294, the horizontal direction is represented by the X
direction which is one spatial direction, and the vertical
direction is represented by the Y direction which is another
spatial direction.
That is to say, the present applicant investigated reproduction
results of a region 3544-1 in the subtraction image 3544 shown in
FIG. 294, which is an example of a region which exhibits small
difference in the pixel value between the HD image 3541 and the
predicted image 3543.
FIG. 295 is a chart which shows: the actual pixel values of the HD
image 3541; the actual pixel values of the SD image 3542,
corresponding to the four pixels from the left side of a series of
six HD pixels in the X direction within the region 3544-1 shown in
FIG. 294; and the actual waveform (signals in the actual world
1).
In FIG. 295, the vertical axis represents the pixel value, and the
horizontal axis represents the x-axis parallel with the spatial
direction X. Note that the X axis is defined with the origin as the
position of the left end of the third HD pixel form the left side
of the six HD pixels within the subtraction image 3544 in the
drawing. Each coordinate value is defined with the origin thus
obtained as the base. Note that the X-axis coordinate values are
defined with the pixel width of an HD pixel of the subtraction
image 3544 as 0.5. That is to say, the subtraction image 3544 is an
HD image, and accordingly, each pixel of the HD image is plotted in
the chart with the pixel width L.sub.t of 0.5 (which will be
referred to as "HD-pixel width L.sub.t" hereafter). On the other
hand, in this case, each pixel of the SD image 3542 is plotted with
the pixel width (which will be referred to as "SD-pixel width
L.sub.s" hereafter) which is twice the HD-pixel width L.sub.t,
i.e., with the SD-pixel width L.sub.s of 1.
Also, in FIG. 295, the solid line represents the pixel values of
the HD image 3541, the dotted line represents the pixel values of
the SD image 3542, and the broken line represents the signal
waveform of the actual world 1 along the X-direction. Note that it
is difficult to plot the actual waveform of the actual world 1 in
reality. Accordingly, the broken line shown in FIG. 295 represents
an approximate function f(x) which approximates the waveform along
the X-direction using the aforementioned linear polynomial
approximation technique (the actual estimating unit 102 according
to the first embodiment shown in FIG. 289).
Then, the present applicant investigated reproduction results in
the region which exhibits large difference in the pixel value
between the HD image 3541 and the predicted image 3543 (i.e., the
region formed with a density close to white, in the subtraction
image 3544) in the same way as in the aforementioned investigation
with regard to the region which exhibits small difference in the
pixel value therebetween. With the aforementioned region,
information obtained for this investigation are: the actual values
of the HD image 3541; the actual pixel values of the SD image 3542;
and the actual waveform corresponding to the HD image 3541 (signals
in the actual world 1), in the same way. The investigation results
are shown in FIG. 296 and FIG. 297.
FIG. 296 shows an example of the investigation-target region. Note
that in FIG. 296, the horizontal direction is represented by the X
direction which is a spatial direction, and the vertical direction
is represented by the Y direction which is another spatial
direction.
That is to say, the present applicant investigated reproduction
results of a region 3544-2 in the subtraction image 3544 shown in
FIG. 296, which is an example of a region which exhibits large
difference in the pixel value between the HD image 3541 and the
predicted image 3543.
FIG. 297 is a chart which shows: the actual pixel values of the HD
image 3541; the actual pixel values of the SD image 3542,
corresponding to the four pixels from the left side of a series of
six HD pixels in the X direction within the region 3544-2 shown in
FIG. 296; and the actual waveform (signals in the actual world
1).
In FIG. 297, the vertical axis represents the pixel value, and the
horizontal axis represents the x-axis parallel with the spatial
direction X. Note that the X axis is defined with the origin as the
position of the left end of the third HD pixel form the left side
of the six HD pixels within the subtraction image 3544 in the
drawing. Each coordinate value is defined with the origin thus
obtained as the base. Note that the X-axis coordinate values are
defined with the SD-pixel width L.sub.s of 1.
In FIG. 297, the solid line represents the pixel values of the HD
image 3541, the dotted line represents the pixel values of the SD
image 3542, and the broken line represents the signal waveform of
the actual world 1 along the X-direction. Note that the broken line
shown in FIG. 297 represents an approximate function f(x) which
approximates the waveform along the X-direction, in the same way as
with the broken line shown in FIG. 295.
Making a comparison between the charts shown in FIG. 295 and FIG.
297, it is clear that each region in the drawing includes the line
object from the waveforms of the approximate functions f(x) shown
in the drawings.
However, there is the difference therebetween as follows. That is
to say, while the line object extends over the region of x of
around 0 to 1 in FIG. 295, the line object extends over the region
of x of around -0.5 to 0.5 in FIG. 297. That is to say, in FIG.
295, the most part of the line object is included within the single
SD pixel positioned at the region of x of 0 to 1 in the SD image
3542. On the other hand, in FIG. 297, a part of the line object is
included within the single SD pixel positioned at the region of x
of 0 to 1 in the SD image 3542 (the edge of the line object
adjacent to the background is also included therewithin).
Accordingly, in a case shown in FIG. 295, there is the small
difference in the pixel value between the two HD pixels
(represented by the solid line) extending the region of x of 0 to
1.0 in the HD image 3541. The pixel value of the corresponding SD
pixel (represented by the dotted line in the drawing) is the
average of the pixel values of the two HD pixels. As a result, it
can be easily understood that there is the small difference in the
pixel value between the SD pixel of the SD image 3542 and the two
HD pixels of the HD image 3541.
In such a state (the state shown in FIG. 295), let us consider
reproduction processing for generating two HD pixels (the pixels of
the predicted image 3543) which extend over the region of x of 0 to
1.0 with the single SD pixel extending the region of x of 0 to 1.0
as the pixel of interest using the conventional class
classification adaptation processing. In this case, the generated
HD pixels of the predicted image 3543 approximate the HD pixels of
the HD image 3541 with sufficiently high precision as shown in FIG.
294. That is to say, in the region 3544-1, there is the small
difference in the pixel value of the HD pixel between the predicted
image 3543 and the HD image 3541, and accordingly, the subtraction
image is formed with a density close to black as shown in FIG.
294.
On the other hand, in a case shown in FIG. 297, there is the large
difference in the pixel value between the two HD pixels
(represented by the solid line) extending the region of x of 0 to
1.0 in the HD image 3541. The pixel value of the corresponding SD
pixel (represented by the dotted line in the drawing) is the
average of the pixel values of the two HD pixels. As a result, it
can be easily understood that there is the large difference in the
pixel value between the SD pixel of the SD image 3541 and the two
HD pixels of the HD image 3541, as compared with the corresponding
difference shown in FIG. 295.
In such a state (the state shown in FIG. 297), let us consider
reproduction processing for generating two HD pixels (the pixels of
the predicted image 3543) which extend over the region of x of 0 to
1.0 with the single SD pixel extending the region of x of 0 to 1.0
as the pixel of interest using the conventional class
classification adaptation processing. In this case, the generated
HD pixels of the predicted image 3543 approximate the HD pixels of
the HD image 3541 with poor precision as shown in FIG. 296. That is
to say, in the region 3544-2, there is the large difference in the
pixel value of the HD pixel between the predicted image 3543 and
the HD image 3541, and accordingly, the subtraction image is formed
with a density close to white as shown in FIG. 296.
Making a comparison between the approximate functions f(x)
(represented by the broken line shown in the drawings) for the
signals in the actual world 1 shown in FIG. 295 and FIG. 297, it
can be understood as follows. That is to say, while the change in
the approximate function f(x) is small over the region of x of 0 to
1 in FIG. 295, the change in the approximate function f(x) is large
over the region of x of 0 to 1 in FIG. 297.
Accordingly, there is an SD pixel in the SD image 3542 as shown in
FIG. 295, which extends over the range of x of 0 to 1.0, over which
the change in the approximate function f(x) is small (i.e., the
change in signals in the actual world 1 is small).
From this perspective, the investigation results described above
can also be said as follows. That is to say, in a case of
reproduction of the HD pixels based upon the SD pixels which
extends over the region over which the change in the approximate
function f(x) is small (i.e., the change in signals in the actual
world 1 is small), such as the SD pixel extending over the region
of x of 0 to 1.0 shown in FIG. 295, using the conventional class
classification adaptation processing, the generated HD pixels
approximate the signals in the actual world 1 (in this case, the
image of the line object) with sufficiently high precision.
On the other hand, there is another SD pixel in the SD image 3542
as shown in FIG. 297, which extends over the range of x of 0 to
1.0, over which the change in the approximate function f(x) is
large (i.e., the change in signals in the actual world 1 is
large).
From this perspective, the investigation results described above
can also be said as follows. That is to say, in a case of
reproduction of the HD pixels based upon the SD pixels which
extends over the region over which the change in the approximate
function f(x) is large (i.e., the change in signals in the actual
world 1 is large), such as the SD pixel extending over the region
of x of 0 to 1.0 shown in FIG. 297, using the conventional class
classification adaptation processing, the generated HD pixels
approximate the signals in the actual world 1 (in this case, the
image of the line object) with poor precision.
The conclusion of the investigation results described above is that
in a case as shown in FIG. 298, it is difficult to reproduce the
details extending over the region corresponding to a single pixel
using the conventional signal processing based upon the relation
between pixels (e.g., the class classification adaptation
processing).
That is to say, FIG. 298 is a diagram for describing the
investigation results obtained by the present applicant.
In FIG. 298, the horizontal direction in the drawing represents the
X-direction which is a direction (spatial direction) along which
the detecting elements of the sensor 2 (FIG. 289) are arrayed. On
the other hand, the vertical direction in the drawing represents
the light-amount level or the pixel value. The dotted line
represents the X cross-sectional waveform F(x) of the signal in the
actual world 1 (FIG. 289). The solid line represents the pixel
value P output from the sensor 2 in a case the sensor 2 receives a
signal (image) in the actual world 1 represented as described
above. Also, the width (length in the X-direction) of a detecting
element of the sensor 2 is represented by L.sub.c. The change in
the X cross-sectional waveform F(x) as to the pixel width L.sub.c
of the sensor 2, which is the width L.sub.c of the detecting
element of the sensor 2, is represented by .DELTA.P.
Here, the aforementioned SD image 3542 (FIG. 293) is an image for
simulating the image (FIG. 289) input from the sensor 2. With this
simulation, evaluation can be made with the SD-pixel width L.sub.s
of the SD image 3542 (FIG. 295 and FIG. 297) as the pixel width
(width of the detecting element) L.sub.c of the sensor 2.
While description has been made regarding investigation for the
signal in the actual world 1 (approximate function f(x)) which
reflects the fine line, there are various types of change in the
signal level in the actual world 1.
Accordingly, the reproduction results under the conditions shown in
FIG. 298 can be estimated based upon the investigation results. The
reproduction results thus estimated are as follows.
That is to say, such as shown in FIG. 298, in a case of reproducing
HD pixels (e.g., pixels of the predicted image output from the
class classification adaptation processing unit 3501 in FIG. 289)
using the conventional class classification adaptation processing
with an SD pixel (output pixel from the sensor 2), over which the
change .DELTA.P in signals in the actual world 1 (the change in the
X cross-sectional waveform F(x)) is large, as the pixel of
interest, the generated HD pixels approximate the signals in the
actual world 1 (X cross-sectional waveform F(x) in a case shown in
FIG. 298) with poor precision.
Specifically, with the conventional methods such as the class
classification adaptation processing, image processing is performed
based upon the relation between multiple pixels output from the
sensor 2.
That is to say, as shown in FIG. 298, let us consider a signal
which exhibits rapid change .DELTA.P in the X cross-sectional
waveform F(x), i.e., rapid change in the signal in the actual world
1, over the region corresponding to a single pixel. Such a signal
is integrated (strictly, time-spatial integration), and only a
single pixel value P is output (the signal over the single pixel is
represented by the uniform pixel value P).
With the conventional methods, image processing is performed with
the pixel value P as both the reference and the target. In other
words, with the conventional methods, image processing is performed
without giving consideration to the change in the signal in the
actual world 1 (X cross-sectional waveform F(x)) over a single
pixel, i.e., without giving consideration to the details extending
over a single pixel.
Any image processing (even class classification adaptation
processing) has difficulty in reproducing change in the signal in
the actual world 1 over a single pixel with high precision as long
as the image processing is performed in increments of pixels. In
particular, great change .DELTA.P in the signal in the actual world
1 leads to marked difficulty therein.
In other words, the problem of the aforementioned class
classification adaptation processing, i.e., in FIG. 289, the cause
of insufficient reproduction of the original details using the
class classification adaptation processing, which often occurs in a
case of employing the input image (SD image) in which the details
have been lost in the stage where the image has been output from
the sensor 2, is as follows. The cause is that the class
classification adaptation processing is performed in increment of
pixels (a single pixel has a single pixel value) without giving
consideration to change in signals in the actual world 1 over a
single pixel.
Note that all the conventional image processing methods including
the class classification adaptation processing have the same
problem, the cause of the problem is completely the same.
As described above, the conventional image processing methods have
the same problem and the same cause of the problem.
On the other hand, the combination of the data continuity detecting
unit 101 and the actual world estimating unit 102 (FIG. 3) allows
estimation of the signals in the actual world 1 based upon the
input image from the sensor 2 (i.e., the image in which the change
in the signal in the actual world 1 has been lost) using the
continuity of the signals in the actual world 1. That is to say,
the actual world estimating unit 102 has a function for outputting
the actual world estimation information which allows estimation of
the signal in the actual world 1.
Accordingly, the change in the signals in the actual world 1 over a
single pixel can be estimated based upon the actual world
estimation information.
In this specification, the present applicant has proposed a class
classification adaptation processing correction method as shown in
FIG. 289, for example, based upon the mechanism in which the
predicted image (which represents the image in the actual world 1,
predicted without giving consideration to the change in the signal
in the actual world 1 over a single pixel) generated by the
conventional class classification adaptation processing is
corrected using a predetermined correction image (which represents
the estimated error of the predicted image due to change in the
signal in the actual world 1 over a single pixel) generated based
on the actual world estimation information, thereby solving the
aforementioned problem.
That is to say, in FIG. 289, the data continuity detecting unit 101
and the actual world estimating unit 102 generate the actual world
estimation information. Then, the class classification adaptation
processing correction unit 3502 generates a correction image having
a predetermined format based upon the actual world estimation
information thus generated. Subsequently, the addition unit 3503
corrects the predicted image output from the class classification
adaptation processing unit 3501 using the correction image output
from the class classification adaptation processing correction unit
3502 (Specifically, makes the sum of the predicted image and the
correction image, and outputs the summed image as an output
image).
Note that detailed description has been made regarding the class
classification adaptation processing unit 3501 included in the
image generating unit 103 for performing class classification
adaptation processing correction method. Also, the type of the
addition unit 3503 is not restricted in particular as long as the
addition unit 3503 has a function of making the sum of the
predicted image and the correction image. Examples employed as the
addition unit 3503 include various types of adders, addition
programs, and so forth.
Accordingly, detailed description will be made below regarding the
class classification adaptation processing correction unit 3502
which has not been described.
First description will be made regarding the mechanism of the class
classification adaptation processing correction unit 3502.
As described above, in FIG. 293, let us assume the HD image 3541 as
the original image (signals in the actual world 1) which is to be
input to the sensor 2 (FIG. 289). Furthermore, let us assume the SD
image 3542 as the input image from the sensor 2. In this case, the
predicted image 3543 can be assumed as the predicted image (image
obtained by predicting the original image (HD image 3541)) output
from the class classification adaptation processing unit 3501.
On the other hand, the image obtained by subtracting the predicted
image 3543 from the HD image 3541 is the subtraction image
3544.
Accordingly, the HD image 3541 is reproduced by actions of: the
class classification adaptation processing correction unit 3502
having a function of creating the subtraction image 3544 and
outputting the subtraction image 3544 as a correction image; and
the addition unit 3503 having a function of making the sum of the
predicted image 3543 output from the class classification
adaptation processing unit 3501 and the subtraction image 3544
(correction image) output from the class classification adaptation
processing correction unit 3502.
That is to say, the class classification adaptation processing
correction unit 3502 suitably predicts the subtraction image (with
the same resolution as with the predicted image output from the
class classification adaptation processing unit 3501), which is the
difference between the image which represents the signals in the
actual world 1 (original image which is to be input to the sensor
2) and the predicted image output from the class classification
adaptation processing unit 3501, and outputs the subtraction image
thus predicted (which will be referred to as "subtraction predicted
image" hereafter) as a correction image, thereby almost completely
reproducing the signals in the actual world 1 (original image).
On the other hand, as described above, there is a relation between:
the difference (error) between the signals in the actual world 1
(the original image which is to be input to the sensor 2) and the
predicted image output from the class classification adaptation
processing unit 3501; and the change in the signals in the actual
world 1 over a single pixel of the input image. Also, the actual
world estimating unit 102 has a function of estimating the signals
in the actual world 1, thereby allowing estimation of the features
for each pixel, representing the change in the signal in the actual
world 1 over a single pixel of the input image.
With such a configuration, the class classification adaptation
processing correction unit 3502 receives the features for each
pixel of the input image, and creates the subtraction predicted
image based thereupon (predicts the subtraction image).
Specifically, for example, the class classification adaptation
processing correction unit 3502 receives an image (which will be
referred to as "feature-amount image" hereafter) from the actual
world estimating unit 102, as the actual world estimation
information in which the features is represented by each pixel
value.
Note that the feature-amount image has the same resolution as with
the input image from the sensor 2. On the other hand, the
correction image (subtraction predicted image) has the same
resolution as with the predicted image output from the class
classification adaptation processing unit 3501.
With such a configuration, the class classification adaptation
processing correction unit 3502 predicts and computes the
subtraction image based upon the feature-amount image using the
conventional class classification adaptation processing with the
feature-amount image as an SD image and with the correction image
(subtraction predicted image) as an HD image, thereby obtaining
suitable subtraction predicted image as a result of the prediction
computation.
The above is the arrangement of the class classification adaptation
processing correction unit 3502.
FIG. 299 shows a configuration example of the class classification
adaptation processing correction unit 3502 which works on the
mechanism.
In FIG. 299, the feature-amount image (SD image) input from the
actual world estimating unit 102 is supplied to region extracting
units 3551 and 3555. The region extracting unit 3551 extracts a
class tap (a set of SD pixels positioned at a predetermined region
including the pixel of interest) necessary for class classification
from the supplied feature-amount image, and outputs the extracted
class tap to a pattern detecting unit 3552. The pattern detecting
unit 3552 detects the pattern of the feature-amount image based
upon the class tap thus input.
A class code determining unit 3553 determines the class code based
upon the pattern detected by the pattern detecting unit 3552, and
outputs the determined class code to correction coefficient memory
3554 and the region extracting unit 3555. The correction
coefficient memory 3554 stores the coefficients for each class
code, obtained by learning. The correction coefficient memory 3554
reads out the coefficients corresponding to the class code input
from the class code determining unit 3553, and outputs the class
code to a correction computing unit 3556.
Note that description will be made later with reference to the
block diagram of the class classification adaptation processing
correction learning unit shown in FIG. 300 regarding the learning
processing for calculating the coefficients stored in the
correction coefficient memory 3554.
On the other hand, the coefficients, i.e., prediction coefficients,
stored in the correction coefficient memory 3554 are used for
predicting the subtraction image (for generating the subtraction
predicted image which is an HD image) as described later. However,
the term, "prediction coefficients" used in the above description
has indicated the coefficients stored in the coefficient memory
3514 (FIG. 290) of the class classification adaptation processing
unit 3501. Accordingly, the prediction coefficients stored in the
correction coefficient memory 3554 will be refereed to as
"correction coefficients" hereafter in order to distinguish the
coefficients from the prediction coefficients stored in the
coefficient memory 3514.
The region extracting unit 3555 extracts a prediction tap (a set of
the SD pixels positioned at a predetermined region including the
pixel of interest) from the feature-amount image (SD image) input
from the actual world estimating unit 102 based upon the class code
input from the class code determining unit 3553, necessary for
predicting the subtraction image (HD image) (i.e., for generating
subtraction predicted image which is an HD image) corresponding to
a class code, and outputs the extracted class tap to the correction
computing unit 3556. The correction computing unit 3556 executes
product-sum computation using the prediction tap input from the
region extracting unit 3555 and the correction coefficients input
from the correction coefficient memory 3554, thereby generating HD
pixels of the subtraction predicted image (HD image) corresponding
to the pixel of interest (SD pixel) of the feature-amount image (SD
image).
More specifically, the correction coefficient memory 3554 outputs
the correction coefficients corresponding to the class code
supplied from the class code determining unit 3553 to the
correction computing unit 3556. The correction computing unit 3556
executes product-sum computation represented by the following
Expression (226) using the prediction tap (SD pixels) extracted
from the pixel values at a predetermined position at a pixel in the
input image supplied from the region extracting unit 3555 and the
correction coefficients supplied from the correction coefficient
memory 3554, thereby obtaining HD pixels of the subtraction
predicted image (HD image) (i.e., predicting and estimating the
subtraction image).
'.times..times. ##EQU00137##
In Expression (226), u' represents the HD pixel of the subtraction
predicted image (HD image). Each of a.sub.i (i represents an
integer of 1 through n) represents the corresponding prediction tap
(SD pixels). On the other hand, each of g.sub.i represents the
corresponding correction coefficient.
Accordingly, while the class classification adaptation processing
unit 3501 shown in FIG. 289 outputs the HD pixel q' represented by
the above Expression (218), the class classification adaptation
processing correction unit 3502 outputs the HD pixel u' of the
subtraction predicted image represented by Expression (226). Then,
the addition unit 3503 makes the sum of the HD pixel q' of the
predicted image and the HD pixel u' of the subtraction predicted
image (which will be represented by "o'" hereafter), and outputs
the sum to external circuits, as an HD pixel of the output
image.
That is to say, the HD pixel o' of the output image output from the
image generating unit 103 in the final stage is represented by the
following Expression (227).
'''.times..times..times..times. ##EQU00138##
FIG. 300 shows a detailed configuration example of the learning
unit for determining the correction coefficients (g.sub.i used in
the above Expression (222)) stored in the correction coefficient
memory 3554 of the class classification adaptation processing
correction unit 3502, i.e., the class classification adaptation
processing correction learning unit 3561 of the learning device
3504 shown in FIG. 291 described above.
In FIG. 291 as described above, upon completion of leaning
processing, the class classification adaptation processing learning
unit 3521 outputs learning predicted image obtained by predicting
the first tutor image based upon the first student image using the
prediction coefficients calculated by learning, as well as
outputting the first tutor image (HD image) and the first student
image (SD image) used for learning processing to the class
classification adaptation processing correction learning unit
3561.
Returning to FIG. 300, of these images, the first student image is
input to a data continuity detecting unit 3572.
On the other hand, of these images, the first tutor image and the
learning predicted image are input to an addition unit 3571. Note
that the learning predicted image is inverted before input to the
addition unit 3571.
The addition unit 3571 makes the sum of the input first tutor image
and the inverted input learning predicted image, i.e., generates a
subtraction image between the first tutor image and the learning
predicted image, and outputs the generated subtraction image to a
normal equation generating unit 3578 as a tutor image used in the
class classification adaptation processing correction learning unit
3561 (which will be referred to as "second tutor image" for
distinguish this image from the first tutor image).
The data continuity detecting unit 3572 detects the continuity of
the data contained in the input first student image, and outputs
the detection results to an actual world estimating unit 3573 as
data continuity information.
The actual world estimating unit 3573 generates a feature-amount
image based upon the data continuity information thus input, and
outputs the generated image to region extracting units 3574 and
3577 as a student image used in the class classification adaptation
processing correction learning unit 3561 (the student image will be
referred to as "second student image" for distinguishing this
student image from the first student image described above).
The region extracting unit 3574 extracts SD pixels (class tap)
necessary for class classification from the second student image
(SD image) thus supplied, and outputs the extracted class tap to a
pattern detecting unit 3575. The pattern detecting unit 3575
detects the pattern of the input class tap, and outputs the
detection results to a class code determining unit 3576. The class
code determining unit 3576 determines the class code corresponding
to the input pattern, and outputs the determined class code to the
region extracting unit 3577 and the normal equation generating unit
3578.
The region extracting unit 3577 extracts the prediction tap (SD
pixels) from the second student image (SD image) input from the
actual world estimating unit 3573 based upon the class code input
from the class code determining unit 3576, and outputs the
extracted prediction tap to the normal equation generating unit
3578.
Note that the aforementioned region extracting unit 3574, the
pattern detecting unit 3575, the class code determining unit 3576,
and the region extracting unit 3577, have generally the same
configurations and functions as with the region extracting unit
3551, the pattern detecting unit 3552, the class code determining
unit 3553, and the region extracting unit 3555 of the class
classification adaptation processing correction unit 3502 shown in
FIG. 299, respectively. Also, the aforementioned data continuity
detecting unit 3572 and the actual world estimating unit 3773 have
generally the same configurations and functions as with the data
continuity detecting unit 101 and the actual world estimating unit
102 shown in FIG. 289, respectively.
The normal equation generating unit 3578 generates a normal
equation based upon the prediction tap (SD pixels) of the second
student image (SD image) input from the region extracting unit 3577
and the HD pixels of the second tutor image (HD image), for each of
the class codes input from the class code determining unit 3576,
and supplies the normal equation to a correction coefficient
determining unit 3579. Upon reception of the normal equation for
the corresponding class code from the normal equation generating
unit 3578, the correction coefficient determining unit 3579
computes the correction coefficients using the normal equation,
which and are stored in the correction coefficient memory 3554 in
association with the class code.
Now, detailed description will be made regarding the normal
equation generating unit 3578 and the correction coefficient
determining unit 3579.
In the above Expression (226), all the correction coefficients
g.sub.i are undetermined before learning. With the present
embodiment, learning is performed by inputting multiple HD pixels
of the tutor image (HD image) for each class code. Let us say that
there are m HD pixels corresponding to a certain class code, and
each of the m HD pixels are represented by u.sub.k (k is an integer
of 1 through m). In this case, the following Expression (228) is
introduced from the above Expression (226).
.times..times..times..times. ##EQU00139##
That is to say, the Expression (228) indicates that the HD pixels
corresponding to a certain class code can be predicted and
estimated by computing the right side of this Expression. Note that
in Expression (228), e.sub.k represents error. That is to say, the
HD pixel U.sub.k' of the subtraction predicted image (HD image)
which is computation results of the right side of this Expression
does not exactly matches the HD pixel u.sub.k of the actual
subtraction image, but contains a certain error e.sub.k.
With Expression (228), the correction coefficients a.sub.i are
obtained by learning such that the sum of squares of the errors
e.sub.k exhibits the minimum, for example.
With the present embodiment, the m (m>n) HD pixels u.sub.k are
prepared for learning processing. In this case, the correction
coefficients a.sub.i can be calculated as a unique solution using
the least square method.
That is to say, the normal equation for calculating the correction
coefficients a.sub.iin the right side of the Expression (228) using
the least square method is represented by the following Expression
(229).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times..times..times..times..-
times..times..times..times..times..times..times..times..times..times..time-
s..times..function.
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times. ##EQU00140##
With the matrix in the Expression (229) as the following
Expressions (230) through (232), the normal equation is represented
by the following Expression (233).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times.
##EQU00141##
As shown in Expression (231), each component of the matrix
G.sub.MAT is the correction coefficient g.sub.i which is to be
obtained. With the present embodiment, in Expression (233), the
matrix A.sub.MAT in the left side thereof and the matrix U.sub.MAT
in the right side thereof are prepared, thereby calculating the
matrix G.sub.MAT (i.e., the correction coefficients g.sub.i) using
the matrix solution method.
Specifically, with the present embodiment, each prediction tap
a.sub.ik is known, and accordingly, each component of the matrix
A.sub.MAT represented by Expression (230) can be obtained. Each
prediction tap a.sub.ik is extracted by the region extracting unit
3577, and the normal equation generating unit 3578 computes each
component of the matrix A.sub.MAT using the prediction tap a.sub.ik
supplied from the region extracting unit 3577.
On the other hand, with the present embodiment, the prediction tap
a.sub.ik and the HD pixel u.sub.k of the subtraction image are
prepared, and accordingly, each component of the matrix U.sub.MAT
represented by Expression (232) can be calculated. Note that the
prediction tap a.sub.ik is the same as that of the matrix
A.sub.MAT. On the other hand, the HD pixel u.sub.k of the
subtraction image matches the corresponding HD pixel of the second
tutor image output from the addition unit 3571. With the present
embodiment, the normal equation generating unit 3578 computes each
component of the matrix U.sub.MAT using the prediction tap a.sub.ik
supplied from the region extracting unit 3577 and the second tutor
image (the subtraction image between the first tutor image and the
learning predicted image).
As described above, the normal equation generating unit 3578
computes each component of the matrix A.sub.MAT and the matrix
U.sub.MAT for each class code, and supplies the computation results
to the correction coefficient determining unit 3579 in association
with the class code.
The correction coefficient determining unit 3579 computes the
correction coefficients g.sub.i each of which is the component of
the matrix G.sub.MAT represented by the above Expression (233)
based upon the normal equation corresponding to the supplied class
code.
Specifically, the normal equation represented by the above
Expression (233) can be transformed into the following Expression
(234).
.times. ##EQU00142##
In Expression (234), each component of the matrix G.sub.MAT in the
left side thereof is the correction coefficient g.sub.i which is to
be obtained. Note that each component of the matrix A.sub.MAT and
each component of the matrix U.sub.MAT are supplied from the normal
equation generating unit 3578. With the present embodiment, upon
reception of the components of the matrix A.sub.MAT in association
with a certain class code and the components of the matrix
U.sub.MAT from the normal equation generating unit 3578, the
correction coefficient determining unit 3579 computes the matrix
G.sub.MAT by executing matrix computation represented by the right
side of Expression (234), and stores the computation results
(correction coefficients g.sub.i) in the correction coefficient
memory 3554 in association with the class code.
The above is the detailed description regarding the class
classification adaptation processing correction unit 3502 and the
class classification adaptation processing correction learning unit
3561 which is a learning unit and a sub-unit of the class
classification adaptation processing correction unit 3502.
Note that the type of the feature-amount image employed in the
present invention is not restricted in particular as long as the
correction image (subtraction predicted image) is generated based
thereupon by actions of the class classification adaptation
processing correction unit 3502. In other words, the pixel value of
each pixel in the feature-amount image, i.e., the features,
employed in the present invention is not restricted in particular
as long as the features represents the change in the signal in the
actual world 1 (FIG. 289) over a single pixel (pixel of the sensor
2 (FIG. 289)).
For example, "intra-pixel gradient" can be employed as the
features.
Note that the "intra-pixel gradient" is a new term defined here.
Description will be made below regarding the intra-pixel
gradient.
As described above, the signal in the actual world 1, which is an
image in FIG. 289, is represented by the function F(x, y, t) with
the positions x, y, and z in the three-dimensional space and time t
as variables.
Now, let us say that the signal in the actual world 1 which is an
image has continuity in a certain spatial direction. In this case,
let us consider a one-dimensional waveform (the waveform obtained
by projecting the function F along the X direction will be referred
to as "X cross-sectional waveform F(x)") obtained by projecting the
function F(x, y, t) along a certain direction (e.g., X-direction)
selected from the spatial directions of the X-direction,
Y-direction, and Z-direction. In this case, it can be understood
that waveforms similar to the aforementioned one-dimensional
waveform F(x) can be obtained therearound along the direction of
the continuity.
Based upon the fact described above, with the present embodiment,
the actual world estimating unit 102 approximates the X
cross-sectional waveform F(x) using a n'th (n represents a certain
integer) polynomial approximate function f(x) based upon the data
continuity information (e.g., angle) which reflects the continuity
of the signal in the actual world 1, which is output form the data
continuity detecting unit 101, for example.
FIG. 301 shows f.sub.4(x) (which is a fifth polynomial function)
represented by the following Expression (235), and f.sub.5(x)
(which is a first polynomial function) represented by the following
Expression (236), for example of such a polynomial approximate
function f(x).
.function..times..times..times..times..times..function.''.times.
##EQU00143##
Note that each of W.sub.0 through W.sub.5 in Expression (235) and
W.sub.0' and W.sub.1' in Expression (236) represents the
coefficient of the corresponding order of the function computed by
the actual world estimating unit 102.
On the other hand, in FIG. 301, the x-axis in the horizontal
direction in the drawing is defined with the left end of the pixel
of interest as the origin (x=0), and represents the relative
position from the pixel of interest along the spatial direction x.
Note that the x-axis is defined with the width L.sub.C of the
detecting element of the sensor 2 as 1. On the other hand, the axis
in the vertical direction in the drawing represents the pixel
value.
As shown in FIG. 301, the one-dimensional approximate function
f.sub.5(x) (approximate function f.sub.5(x) represented by
Expression (232)) approximates the X cross-sectional waveform F(x)
around the pixel of interest using collinear approximation. In this
specification, the gradient of the linear approximate function will
be referred to as "intra-pixel gradient". That is to say, the
intra-pixel gradient is represented by the coefficient w.sub.1' of
x in Expression (236).
The rapid intra-pixel gradient reflects great change in the X
cross-sectional waveform F(x) around the pixel of the interest. On
the other hand, the gradual gradient reflects small change in the X
cross-sectional waveform F(x) around the pixel of interest.
As described above, the intra-pixel gradient suitably reflects
change in the signal in the actual world 1 over a single pixel
(pixel of the sensor 2). Accordingly, the intra-pixel gradient may
be employed as the features.
For example, FIG. 302 shows the actual feature-amount image
generated with the intra-pixel gradient as the features.
That is to say, the image on the left side in FIG. 302 is the same
as the SD image 3542 shown in FIG. 293 described above. On the
other hand, the image on the right side in FIG. 302 is a
feature-amount image 3591 generated as follows. That is to say, the
intra-pixel gradient is obtained for each pixel of the SD image
3542 on the left side in the drawing. Then, the image on the right
side in the drawing is generated with the value corresponding to
the intra-pixel gradient as the pixel value. Note that the
feature-amount image 3591 has the nature as follows. That is to
say, in a case of the intra-pixel gradient of zero (the linear
approximate function is parallel with the X-direction), the image
is generated with a density corresponding to black. On the other
hand, in a case of the intra-pixel gradient of 90.degree. (the
linear approximate function is parallel with the Y-direction), the
image is generated with a density corresponding to white.
The region 3542-1 in the SD image 3542 corresponds to the region
3544-1 (which has been used in the above description with reference
to FIG. 295, as an example of the region in which change in the
signal in the actual world 1 is small over a single pixel) in the
subtraction image 3544 shown in FIG. 294 described above. The
region of the feature-amount image 3591 corresponding to the region
3542-1 in the SD image 3542 is the region 3591-1.
On the other hand, the region 3542-2 in the SD image 3542
corresponds to the region 3544-2 (which has been used in the above
description with reference to FIG. 297, as an example of the region
in which change in the signal in the actual world 1 is large over a
single pixel) in the subtraction image 3544 shown in FIG. 296
described above. The region of the feature-amount image 3591
corresponding to the region 3542-2 in the SD image 3542 is the
region 3591-2.
Making a comparison between the region 3542-1 of the SD image 3542
and the region 3591-1 of the feature-amount image 3591, it can be
understood that the region in which change in the signal in the
actual world 1 is small corresponds to the region of the
feature-amount image 3591 having a density close to black
(corresponding to the region having a gradual intra-pixel
gradient).
On the other hand, making a comparison between the region 3542-2 of
the SD image 3542 and the region 3591-2 of the feature-amount image
3591, it can be understood that the region in which change in the
signal in the actual world 1 is large corresponds to the region of
the feature-amount image 3591 having a density close to white
(corresponding to the region having a rapid intra-pixel
gradient).
As described above, the feature-amount image generated with the
value corresponding to the intra-pixel gradient as the pixel value
suitably reflects the degree of change in the signal in the actual
world 1 for each pixel.
Next, description will be made regarding a specific computing
method for the intra-pixel gradient.
That is to say, with the intra-pixel gradient around the pixel of
interest as "grad", the intra-pixel gradient grad is represented by
the following Expression (237).
' ##EQU00144##
In Expression (237), P.sub.n represents the pixel value of the
pixel of interest. Also, P.sub.C represents the pixel value of the
center pixel.
Specifically, as shown in FIG. 303, let us consider a region 3601
(which will be referred to as "continuity region 3601" hereafter)
of 5.times.5 pixels (square region of 5.times.5=25 pixels in the
drawing) in the input image from the sensor 2, having a certain
data continuity. In a case of the continuity region 3601, the
center pixel is the pixel 3602 positioned at the center of the
continuity region 3601. Accordingly, P.sub.C is the pixel value of
the center pixel 3602. Also, in a case that the pixel 3603 is the
pixel of interest, P.sub.n is the pixel value of the pixel of
interest 3603.
Also, in Expression (237), x.sub.n' represents the cross-sectional
direction distance at the center of the pixel of interest. Note
that with the center of the center pixel (pixel 3602 in a case
shown in FIG. 303) as the origin (0, 0) in the spatial directions,
"the cross-sectional direction distance" is defined as the relative
distance along the X-direction between the center pixel of interest
and the line (the line 3604 in a case shown in FIG. 303) which is
parallel with the data-continuity direction, and which passes
through the origin.
FIG. 304 is a diagram which shows the cross-sectional direction
distance for each pixel within the continuity region 3601 in FIG.
303. That is to say, in FIG. 304, the value marked within each
pixel in the continuity region 3601 (square region of 5.times.5=25
pixels in the drawing) represents the cross-sectional direction
distance at the corresponding pixel. For example, the
cross-sectional direction distance X.sub.n' at the pixel of
interest 3603 is -2.beta..
Note that the X-axis and the Y-axis are defined with the pixel
width of 1 in both the X-direction and the Y-direction.
Furthermore, the X-direction is defined with the positive direction
matching the right direction in the drawing. Also, in this case,
.beta. represents the cross-sectional direction distance at the
pixel 3605 adjacent to the center pixel 3602 in the Y-direction
(adjacent thereto downward in the drawing). With the present
embodiment, the data continuity detecting unit 101 supplies the
angle .theta. (the angle .theta. between the direction of the line
3604 and the X-direction) as shown in FIG. 304 as the data
continuity information, and accordingly, the value .beta. can be
obtained with ease using the following Expression (238).
.beta..times..times..theta. ##EQU00145##
As described above, the intra-pixel gradient can be obtained with
simple computation based upon the two input pixel values of the
center pixel (e.g., pixel 3602 in FIG. 304) and the pixel of
interest (e.g., pixel 3603 in FIG. 304) and the angle .theta.. With
the present embodiment, the actual world estimating unit 102
generates a feature-amount image with the value corresponding to
the intra-pixel gradient as the pixel value, thereby greatly
reducing the processing amount.
Note that with an arrangement which requires higher-precision
intra-pixel gradient, the actual-world estimating unit 102 should
compute the intra-pixel gradient using the pixels around and
including the pixel of interest with the least square method.
Specifically, let us say that m (m represents an integer of 2 or
more) pixels around and including the pixel of interest are
represented by index number i (i represents an integer of 1 through
m). The actual world estimating unit 102 substitutes the input
pixel values P.sub.i and the corresponding cross-sectional
direction distance x.sub.i' into the right side of the following
Expression (239), thereby computing the intra-pixel gradient grad
at the pixel of interest. That is to say, Expression (239) is the
same expression as the above expression for obtaining one variable
with the least square method.
.times..times..times.' ##EQU00146##
Next, description will be made with reference to FIG. 305 regarding
processing (processing in Step S103 shown in FIG. 40) for
generating an image performed by the image generating unit 103
(FIG. 289) using the class classification adaptation processing
correction method.
In FIG. 289, upon reception of the signal in the actual world 1
which is an image, the sensor 2 outputs the input image. The input
image is input to the class classification adaptation processing
unit 3501 of the image generating unit 103 as well as being input
to the data continuity detecting unit 101.
Then, in Step S3501 shown in FIG. 305, the class classification
adaptation processing unit 3501 performs class classification
adaptation processing for the input image (SD image) so as to
generate the predicted image (HD image), and outputs the generated
predicted image to the addition unit 3503.
Note that such processing in Step S3501 performed by the class
classification adaptation processing unit 3501 will be referred to
as "input image class classification adaptation processing"
hereafter. Detailed description will be made later with reference
to the flowchart shown in FIG. 306 regarding the "input image class
classification adaptation processing" in this case.
The data continuity detecting unit 101 detects the data continuity
contained in the input image at almost the same time as with the
processing in Step S3501, and outputs the detection results (angle
in this case) to the actual world estimating unit 102 as data
continuity information (processing in Step S101 shown in FIG.
40).
The actual world estimating unit 102 generates the actual world
estimation information (the feature-amount image which is an SD
image in this case) based upon the input angle (data continuity
information), and supplies the actual world estimation information
to the class classification adaptation processing correction unit
3502 (processing in Step S102 shown in FIG. 40).
Then, in Step S3502, the class classification adaptation processing
correction unit 3502 performs class classification adaptation
processing for the feature-amount image (SD image) thus supplied,
so as to generate the subtraction predicted image (HD image) (i.e.,
so as to predict and compute the subtraction image (HD image)
between the actual image (signal in the actual world 1) and the
predicted image output from the class classification adaptation
processing unit 3501), and outputs the subtraction predicted image
to the addition unit 3503 as a correction image.
Note that such processing in Step S3502 performed by the class
classification adaptation processing correction unit 3502 will be
referred to as "class classification adaptation processing
correction processing" hereafter. Detailed description will be made
later with reference to the flowchart shown in FIG. 307 regarding
the "class classification adaptation processing correction
processing" in this case.
Then, in Step S3503, the addition unit 3503 makes the sum of: the
pixel of interest (HD pixel) of the predicted image (HD image)
generated with the processing shown in Step S3501 by the class
classification adaptation processing unit 3501; and the
corresponding pixel (HD pixel) of the correction image (HD image)
generated with the processing shown in Step S3502 by the class
classification adaptation processing correction unit 3502, thereby
generating the pixel (HD pixel) of the output image (HD pixel).
In Step S3504, the addition unit 3503 determines whether or not the
processing has been performed for all the pixels.
In the event that determination has been made that the processing
has not been performed for all the pixels in Step S3504, the
processing returns to Step S3501, and the subsequent processing is
repeated. That is to say, the processing in Steps S3501 through
S3503 is repeated for each of the remaining pixels which have not
been taken as a pixel of interest so as to be taken as a pixel of
interest in order.
Upon completion of the processing for all the pixels (in the event
that determination has been made that processing has been performed
for all the pixels in Step S3504), the addition unit 3504 outputs
the output image (HD image) to external circuits in Step S3505,
whereby processing for generating an image ends.
Next, detailed description will be made with reference to the
drawings regarding the "input image class classification adaptation
processing (the processing in Step S3501)", and the "class
classification adaptation correction processing (the processing in
Step S3502)", step by step in that order.
First, detailed description will be made with reference to the
flowchart shown in FIG. 306 regarding the "input image class
classification adaptation processing" executed by the class
classification adaptation processing unit 3501 (FIG. 290).
Upon input of the input image (SD image) to the class
classification adaptation processing unit 3501, the region
extracting units 3511 and 3515 each receive the input image in Step
S3521.
In Step S3522, the region extracting unit 3511 extracts the pixel
of interest (SD pixel) from the input image and (one or more)
pixels (SD pixels) at predetermined relative positions away from
the pixel of interest as a class tap, and supplies the extracted
class tap to the pattern detecting unit 3512.
In Step S3523, the pattern detecting unit 3512 detects the pattern
of the class tap thus supplied, and supplies the detected pattern
to the class code determining unit 3513.
In Step S3524, the class code determining unit 3513 determines the
class code suited to the pattern of the class tap thus supplied,
from the multiple class codes prepared beforehand, and supplies the
determined class code to the coefficient memory 3514 and the region
extracting unit 3515.
In Step S3525, the coefficient memory 3514 selects the prediction
coefficients (set) corresponding to the supplied class code, which
are to be used in the subsequent processing, from the multiple
prediction coefficients (set) determined beforehand with learning
processing, and supplies the selected prediction coefficients to
the prediction computing unit 3516.
Note that description will be made later regarding the learning
processing with reference to the flowchart shown in FIG. 311.
In Step S3526, the region extracting unit 3515 extracts the pixel
of interest (SD pixel) from the input image and (one or more)
pixels (SD pixels) at predetermined relative positions (which may
be set to the same positions as with the class tap) away from the
pixel of interest as a prediction tap, and supplies the extracted
prediction tap to the prediction computing unit 3516.
In Step S3527, the prediction computing unit 3516 performs
computation processing for the prediction tap supplied from the
region extracting unit 3515 using the prediction coefficients
supplied from the coefficient memory 3514 so as to generate the
predicted image (HD image), and outputs the generated predicted
image to the addition unit 3503.
Specifically, the prediction computing unit 3516 performs
computation processing as follows. That is to say, with each pixel
of the prediction tap supplied from the region extracting unit 3515
as c.sub.i (i represents an integer of 1 through n), and with each
of the prediction coefficients supplied from the coefficient memory
3514 as d.sub.i, the prediction computing unit 3516 performs
computation represented by the right side of the above Expression
(218), thereby calculating the HD pixel q' corresponding to the
pixel of interest (SD pixel). Then, the prediction computing unit
3516 outputs the calculated HD pixel q' to the addition unit 3503
as a pixel forming the predicted image (HD image), whereby the
input image class classification adaptation processing ends.
Next, detailed description will be made with reference to the
flowchart shown in FIG. 307 regarding the "class classification
adaptation processing correction processing" executed by the class
classification adaptation processing correction unit 3502 (FIG.
299).
Upon input of the feature-amount image(SD image) to the class
classification adaptation processing correction unit 3502 as the
actual world estimation information from the actual world
estimating unit 102, the region extracting units 3551 and 3555 each
receive the feature-amount image in Step S3541.
In Step S3542, the region extracting unit 3551 extracts the pixel
of interest (SD pixel) and (one or more) pixels (SD pixels) at
predetermined relative positions away from the pixel of interest
from the feature amount image as a class tap, and supplies the
extracted class tap to the pattern detecting unit 3552.
Specifically, in this case, let us say that the region extracting
unit 3551 extracts a class tap (a set of pixels) 3621 shown in FIG.
308, for example. That is to say, FIG. 308 shows an example of the
layout of the class tap.
In FIG. 308, the horizontal axis in the drawing represents the
X-direction which is one spatial direction, and the vertical
direction in the drawing represents the Y-direction which is
another spatial direction. Note that the pixel of interest is
represented by the pixel 3621-2.
In this case, with the example shown in FIG. 308, the pixels
extracted as the class tap are a total of five pixels of: the pixel
of interest 3621-1; the pixels 3621-0 and 3621-4 which are adjacent
to the pixel of interest 3621-2 along the Y-direction; and the
pixels 3621-1 and 3621-3 which are adjacent to the pixel of
interest 3621-2 along the X-direction, which make up a pixel set
3621.
It is needless to say that the layout of the class tap employed in
the present embodiment is not restricted to the example shown in
FIG. 308, rather, various kinds of layouts may be employed as long
as it includes the pixel of interest 3624-2.
Returning to FIG. 307, in Step S3543, the pattern detecting unit
3552 detects the pattern of the class tap thus supplied, and
supplies the detected pattern to the class code determining unit
3553.
Specifically, in this case, the pattern detecting unit 3552 detects
the class which belongs the pixel value, i.e., the value of
features (e.g., intra-pixel gradient), for each of the five pixels
3621-0 through 3621-4 forming the class tap shown in FIG. 308, and
outputs the detection results in the form of a single data set as a
pattern, for example.
Now, let us say that a pattern shown in FIG. 309 is detected, for
example. That is to say, FIG. 309 shows an example of the pattern
of the class tap.
In FIG. 309, the horizontal axis in the drawing represents the
class taps, and the vertical axis in the drawing represents the
intra-pixel gradient. On the other hand, let us say that the
classes prepared beforehand are a total of three classes of class
3631, class 3632, and class 3633.
In this case, FIG. 309 shows a pattern in which the class tap
3621-0 belongs the class 3631, the class tap 3621-1 belongs the
class 3631, the class tap 3621-2 belongs the class 3633, the class
tap 3621-3 belongs the class 3631, and the class tap 3621-4 belongs
the class 3632.
As described above, each of the five class taps 3621-0 through
3621-4 belongs to one of the three classes 3631 through 3633.
Accordingly, in this case, there are a total of 273 (=3^5) patterns
including the pattern shown in FIG. 309.
Returning to FIG. 307, in Step S3544, the class code determining
unit 3553 determines the class code corresponding to the pattern of
the class tap thus supplied, from multiple class code prepared
beforehand, and supplies the determined class code to the
correction coefficient memory 3554 and the region extracting unit
3555. In this case, there are 273 patterns, and accordingly, there
are 273 (or more) class codes prepared beforehand.
In step S3545, the correction coefficient memory 3554 selects the
correction coefficients (set), which are to be used in the
subsequent processing, corresponding to the class code thus
supplied, from the multiple sets of the correction coefficient set
determined beforehand with the learning processing, and supplies
the selected correction coefficients to the correction computing
unit 3556. Note that each of the correction-coefficient sets
prepared beforehand is stored in the correction coefficient memory
3554 in association with one of the class codes prepared
beforehand. Accordingly, in this case, the number of the
correction-coefficient sets matches the number of the class codes
prepared beforehand (i.e., 273 or more).
Note that description will be made later regarding the learning
processing with reference to the flowchart shown in FIG. 311.
In Step S3546, the region extracting unit 3555 extracts the pixel
of interest (SD pixel) from the input image and the pixels (SD
pixels) at predetermined relative positions (One or more positions
determined independent of those of the class taps. However, the
positions of the prediction tap may match those of the class tap)
away from the pixel of interest, which are used as class taps, and
supplies the extracted prediction taps to the correction computing
unit 3556.
Specifically, in this case, let us say that the prediction tap
(set) 3641 shown in FIG. 310 is extracted. That is to say, FIG. 310
shows an example of the layout of the prediction tap.
In FIG. 310, the horizontal axis in the drawing represents the
X-direction which is one spatial direction, and the vertical
direction in the drawing represents the Y-direction which is
another spatial direction. Note that the pixel of interest is
represented by the pixel 3641-1. That is, the pixel 3641-1 is a
pixel corresponding to the class tap 3621-2 (FIG. 308).
In this case, with the example shown in FIG. 310, the pixels
extracted as the prediction tap (group) are 5.times.5 pixels 3041
(a set of pixels formed of a total of 25 pixels) with the pixel of
interest 3641-1 as the center.
It is needless to say that the layout of the prediction tap
employed in the present embodiment is not restricted to the example
shown in FIG. 310, rather, various kinds of layouts including the
pixel of interest 3641-1 may be employed.
Returning to FIG. 307, in Step S3547, the correction computing unit
3556 performs computation for the prediction taps supplied from the
region extracting unit 3555 using the prediction coefficients
supplied from the correction coefficient memory 3554, thereby
generating subtraction predicted image (HD image). Then, the
correction computing unit 3556 outputs the subtraction predicted
image to the addition unit 3503 as a correction image.
More specifically, with each of the class taps supplied from the
region extracting unit 3555 as a.sub.i (i represents an integer of
1 through n), and with each of the correction coefficients supplied
from the correction coefficient memory 3554 as g.sub.i, the
correction computing unit 3556 performs computation represented by
the right side of the above Expression (226), thereby calculating
the HD pixel u' corresponding to the pixel of interest (SD pixel).
Then, the correction computing unit 3556 outputs the calculated HD
pixel to the addition unit 3503 as a pixel of the correction image
(HD image), whereby the class classification adaptation correction
processing ends.
Next, description will be made with reference to the flowchart
shown in FIG. 311 regarding the learning processing performed by
the learning device (FIG. 291), i.e., the learning processing for
generating the prediction coefficients used in the class
classification adaptation processing unit 3501 (FIG. 290), and the
learning processing for generating the correction coefficients used
in the class classification adaptation processing correction unit
3502 (FIG. 299).
In Step S3561, the class classification adaptation processing
learning unit 3521 generates the prediction coefficients used in
the class classification adaptation processing unit 3501.
That is to say, the class classification adaptation processing
learning unit 3521 receives a certain image as a first tutor image
(HD image), and generates a student image (SD image) with a reduced
resolution based upon the first tutor image.
Then, the class classification adaptation processing learning unit
3521 generates the prediction coefficients which allows suitable
prediction of the first tutor image (HD image) based upon the first
student image (SD image) using the class classification adaptation
processing, and stores the generated prediction coefficients in the
coefficient memory 3514 (FIG. 290) of the class classification
adaptation processing unit 3501.
Note that such processing shown in Step S3561 executed by the class
classification adaptation processing learning unit 3521 will be
referred to as "class classification processing learning
processing" hereafter. Detailed description will be made later
regarding the "class classification adaptation processing learning
unit" in this case, with reference to the flowchart shown in FIG.
312.
Upon generation of the prediction coefficients used in the class
classification adaptation processing unit 3501, the class
classification adaptation processing correction learning unit 3561
generates the correction coefficients used in the class
classification adaptation processing correction unit 3502 in Step
S3562.
That is to say, the class classification adaptation processing
correction learning unit 3561 receives the first tutor image, the
first student image, and the learning predicted image (the image
obtained by predicting the first tutor image using the prediction
coefficients generated by the class classification adaptation
processing learning unit 3521), from the class classification
adaptation processing learning unit 3521.
Next, the class classification adaptation processing correction
learning unit 3561 generates the subtraction image between the
first tutor image and the learning predicted image, which is used
as the second tutor image, as well as generating the feature-amount
image based upon the first student image, which is used as the
second student image.
Then, the class classification adaptation processing correction
learning unit 3561 generates prediction coefficients which allow
suitable prediction of the second tutor image (HD image) based upon
the second student image (SD image) using the class classification
adaptation processing, and stores the generated prediction
coefficients in the correction coefficient memory 3554 of the class
classification adaptation processing correction unit 3502 as the
correction coefficients, whereby the learning processing ends.
Note that such processing shown in Step S3562 executed by the class
classification adaptation processing correction learning unit 3561
will be referred to as "class classification adaptation processing
correction learning processing" hereafter. Detailed description
will be made later regarding the "class classification adaptation
processing correction learning processing" in this case, with
reference to the flowchart shown in FIG. 313.
Next, description will be made regarding "class classification
adaptation processing learning processing (processing in Step
S3561)" and "class classification adaptation processing correction
learning processing (processing in Step S3562)" in this case, step
by step in that order, with reference to the drawings.
First, detailed description will be made with reference to the
flowchart shown in FIG. 312 regarding the "class classification
adaptation processing learning processing" executed by the class
classification adaptation processing learning unit 3521 (FIG.
292).
In Step S3581, the down-converter unit 3531 and the normal equation
generating unit 3536 each receive a certain image as the first
tutor image (HD image). Note that the first tutor image is also
input to the class classification adaptation processing correction
learning unit 3561, as described above.
In Step S3582, the down-converter unit 3531 performs
"down-converting" processing (image conversion into a
reduced-resolution image) for the input first tutor image, thereby
generating the first student image (SD image). Then, the
down-converter unit 3531 supplies the generated first student image
to the class classification adaptation processing correction
learning unit 3561, as well as to the region extracting units 3532
and 3535.
In Step S3583, the region extracting unit 3532 extracts the class
taps from the first student image thus supplied, and outputs the
extracted class taps to the pattern detecting unit 3533. While
strictly, there is the difference (such difference will be referred
to simply as "difference in input/output" hereafter) in the
input/output of information to/from a block between the processing
shown in Step S3583 and the aforementioned processing shown in Step
S3522 (FIG. 306), the processing shown in Step S3583 is generally
the same as that shown in Step S3522 described above.
In Step S3584, the pattern detecting unit 3533 detects the pattern
from the supplied class taps for determining the class code, and
supplies the detected pattern to the class code determining unit
3534. Note that the processing shown in Step S3584 is generally the
same as that shown in Step S3523 (FIG. 306) described above, except
for input/output.
In Step S3585, the class code determining unit 3534 determines the
class code based upon the pattern of the class taps thus supplied,
and supplies the determined class code to the region extracting
unit 3535 and the normal equation generating unit 3536. Note that
the processing shown in Step S3585 is generally the same as that
shown in Step S3524 (FIG. 306) described above, except for
input/output.
In Step S3586, the region extracting unit 3535 extracts the
prediction taps from the first student image corresponding to the
supplied class code, and supplies the extracted prediction taps to
the normal equation generating unit 3536 and the prediction
computing unit 3538. Note that the processing shown in Step S3586
is generally the same as that shown in Step S3526 (FIG. 306)
described above, except for input/output.
In Step S3587, the normal equation generating unit 3536 generates a
normal equation represented by the above Expression (220) (i.e.,
Expression (221)) based upon the prediction taps (SD pixels)
supplied from the region extracting unit 3535 and the corresponding
HD pixels of the HD pixels of the first tutor image (HD image), and
supplies the generated normal equation to the coefficient
determining unit 3537 along with the class code supplied from the
class code determining unit 3534.
In Step S3588, the coefficient determining unit 3537 solves the
normal equation thus supplied, thereby determining the prediction
coefficients. That is to say, the coefficient determining unit 3537
computes the right side of the above Expression (225), thereby
calculating the prediction coefficients. Then, the coefficient
determining unit 3537 supplies the determined prediction
coefficients to the prediction computing unit 3538, as well as
storing the prediction coefficients in the coefficient memory 3514
in association with the class code thus supplied.
In Step S3589, the prediction computing unit 3538 performs
computation for the prediction taps supplied from the region
extracting unit 3535 using the prediction coefficient supplied from
the coefficient determining unit 3537, thereby generating the
learning predicted image (HD pixels).
Specifically, with each of the prediction taps supplied from the
region extracting unit 3535 as c.sub.i (i represents an integer of
1 through n), and with each of the prediction coefficients supplied
from the coefficient determining unit 3537 as d.sub.i, the
prediction computing unit 3538 computes the right side of the above
Expression (218), thereby calculating an HD pixel q' which is
employed as a pixel of the learning predicted image, and which
predicts the corresponding HD pixel q of the first tutor image.
In Step S3590, determination has been made whether or not such
processing has been performed for all the pixels. In the event that
determination has been made that the processing has not been
performed for all the pixels, the flow returns to Step S3583. That
is to say, the processing in Step S3533 through 3590 is repeated
until completion of the processing for all the pixels.
Then, in Step S3590, in the event that determination has been made
that the processing is performed for all the pixels, the prediction
computing unit 3538 outputs the learning predicted image (HD image
formed of the HD pixels q' each of which has been generated for
each processing in Step S3589) to the class classification
adaptation processing correction learning unit 3561, whereby the
class classification adaptation processing learning processing
ends.
As described above, in this example, following completion of the
processing for all the pixels, the learning predicted image which
is an HD image that predicts the first tutor image is input to the
class classification adaptation processing correction learning unit
3561. That is to say, all the HD pixels (predicted pixels) forming
an image is output at the same time.
However, the present invention is not restricted to the
aforementioned arrangement in which all the pixels forming an image
are output at the same. Rather, an arrangement may be made in which
the generated HD pixel is output to the class classification
adaptation processing correction learning unit 3561 each time that
the HD pixel (predicted pixel) is generated by the processing in
Step S3589. With such an arrangement, the processing in Step S3591
is omitted.
Next, detailed description will be made with reference to the
flowchart shown in FIG. 313 regarding "class classification
adaptation processing correction learning processing" executed by
the class classification adaptation processing correction learning
unit 3561 (FIG. 300).
Upon reception of the first tutor image (HD image) and the learning
predicted image (HD image) from the class classification adaptation
processing learning unit 3521, in Step S3601, the addition unit
3571 subtracts the learning predicted image from the first tutor
image, thereby generating the subtraction image (HD image). Then,
the addition unit 3571 supplies the generated subtraction image to
the normal equation generating unit 3578 as the second tutor
image.
Upon reception of the first student image (SD image) from the class
classification adaptation processing learning unit 3521, in Step
S3602, the data continuity detecting unit 3572 and the actual world
estimating unit 3573 generate the feature-amount image based upon
the input first student image (SD image), and supply the generated
feature-amount image to the region extracting units 3574 and 3577
as the second student image.
That is to say, the data continuity detecting unit 3572 detects the
data continuity contained in the first student image, and outputs
the detection results (angle, in this case) to the actual world
estimating unit 3573 as data continuity information. Note that the
processing shown in Step S3602 performed by the data continuity
detecting unit 3572 is generally the same as that shown in Step
S101 shown in FIG. 40 described above, except for input/output.
The actual world estimating unit 3573 generates the actual world
estimation information (feature-amount image which is an SD image,
in this case) based upon the angle (data continuity information)
thus input, and supplies the generated actual world estimation
information to the region extracting unit 3574 and 3577 as the
second student image. Note that the processing shown in Step S3602
performed by the actual world estimating unit 3573 is generally the
same as that shown in Step S102 shown in FIG. 40 described above,
except for input/output.
Note that the present invention is not restricted to an arrangement
in which the processing in Step S3601 and the processing in Step
S3602 are performed in that order shown in FIG. 313. That is to
say, an arrangement may be made in which the processing in Step
S3602 is performed upstream the processing in Step S3601.
Furthermore, the processing in Step S3601 and the processing in
Step S3602 may be performed at the same time.
In Step S3603, the region extracting unit 3574 extracts the class
taps from the second student image (feature-amount image) thus
supplied, and outputs the extracted class taps to the pattern
detecting unit 3575. Note that the processing shown in Step S3603
is generally the same as that shown in Step S3542 (FIG. 307)
described above, except for input/output. That is to say, in this
case, a set of pixels 3621 having a layout shown in FIG. 308 is
extracted as class taps.
In Step S3604, the pattern detecting unit 3575 detects the pattern
from the class taps thus supplied for determining the class code,
and supplies the detected pattern to the class code determining
unit 3576. Note that the processing shown in Step S3604 is
generally the same as that shown in Step S3543 (FIG. 307) described
above, except for input/output. That is to say, in this case, the
pattern detecting unit 3575 detects at least 273 patterns at the
time of completion of the learning processing.
In Step S3605, the class code determining unit 3576 determines the
class code based upon the pattern of the class taps thus supplied,
and supplies the class code to the region extracting unit 3577 and
the normal equation generating unit 3578. Note that the processing
shown in Step S3605 is generally the same as that shown in Step
S3544 (FIG. 307) described above, except for input/output. That is
to say, in this case, the class code determining unit 3576
determines at least 273 class codes at the time of completion of
the learning processing.
In Step S3606, the region extracting unit 3577 extracts the
prediction taps corresponding to the class code thus supplied, from
the second student image (feature-amount image), and supplies the
extracted prediction taps to the normal equation generating unit
3578. Note that the processing shown in Step S3606 is generally the
same as that shown in Step S3546 (FIG. 307) described above, except
for input/output. That is to say, in this case, a set of pixels 354
having a layout shown in FIG. 310 is extracted as prediction
taps.
In step S3607, the normal equation generating unit 3578 generates a
normal equation represented by the above Expression (229) (i.e.,
Expression (230)) based upon the prediction taps (SD pixels)
supplied from the region extracting unit 3577 and the second tutor
image (subtraction image between the first tutor image and the
learning predicted image, which is an HD image), and supplies the
generated normal equation to the correction coefficient determining
unit 3579 along with the class code supplied from the class code
determining unit 3576.
In Step S3608, the correction coefficient determining unit 3579
determines the correction coefficients by solving the normal
equation thus supplied, i.e., calculates the correction
coefficients by computing the right side of the above Expression
(234), and stores the calculated correction coefficients associated
with the supplied class code in the correction coefficient memory
3554.
In Step S3609, determination is made whether or not such processing
has been performed for all the pixels. In the event that
determination has been made that the processing has not been
performed for all the pixels, the flow returns to Step S3603. That
is to say, the processing in Step S3603 through 3609 is repeated
until completion of the processing for all the pixels.
On the other hand, in Step S3609, in the event that determination
has been made that the processing has been performed for all the
pixels, the class classification adaptation processing correction
learning processing ends.
As described above, with the class classification adaptation
correction processing method, the summed image is generated by
making the sum of the predicted image output from the class
classification adaptation processing unit 3501 and the correction
image (subtraction predicted image) output from the class
classification adaptation processing correction unit 3502, and the
summed image thus generated is output.
For example, let us say that the HD image 3541 shown in FIG. 293
described above is converted to a reduced-resolution image, i.e.,
the SD image 3542 with a reduced resolution is obtained, and the SD
image 3542 thus obtained is employed as an input image. In this
case, the class classification adaptation processing unit 3501
outputs the predicted image 3543 shown in FIG. 314. Then, the
summed image is generated by making the sum of the predicted image
3543 and the correction image (not shown) output from the class
classification adaptation processing correction unit 3502 (e.g.,
the predicted image 3543 is corrected using the correction image),
thereby generating the output image 3651 shown in FIG. 294.
Making a comparison between the output image 3651, the predicted
image 3543, and the HD image 3541 (FIG. 293) which is an original
image, it has been confirmed that the output image 3651 is more
similar to the HD image 3541 than the predicted image 3543.
As described above, the class classification adaptation processing
correction method enables output of an image more similar to the
original image (the signal in the actual world 1 which is to be
input to the sensor 2), in comparison with other techniques
including class classification adaptation processing.
In other words, with the class classification adaptation processing
correction method, for example, the data continuity detecting unit
101 shown in FIG. 289 detects the data continuity contained in the
input image (FIG. 289) formed of multiple pixels having the pixel
values obtained by projecting the light signals in the actual world
1 shown in FIG. 289 by actions of multiple detecting elements of a
sensor (e.g., the sensor 2 shown in FIG. 289), in which a part of
the continuity as the light signals in the actual world has been
lost due to the projection of the light signals in the actual world
1 to the pixel values by actions of the multiple detecting elements
each of which has the nature of time-spatial integration
effects.
For example, the actual world estimating unit 102 shown in FIG. 289
detects the actual world feature contained in the light-signal
function F(x) (FIG. 298) which represents the light signals of the
actual world 1 (e.g., the features corresponding to the pixel of
the feature-amount image shown in FIG. 289), corresponding to the
detected data continuity, thereby estimating the light signals in
the actual world 1.
Specifically, for example, making an assumption that the pixel
value which represents the distance (e.g., the cross-sectional
direction distance Xn' shown in FIG. 303) from the line (e.g., the
line 3604 in FIG. 303), which represents the data continuity thus
supplied, along at least one dimensional direction represents the
at least one-dimensional integration effects which have affected
the corresponding pixel, the actual world estimating unit 102
approximates the light-signal function F(x) with the approximate
function f.sub.5(x) shown in FIG. 301, for example, and detects the
intra-pixel gradient (e.g., grad in the above Expression (234), and
the coefficient w1' of x in Expression (233)) which is the gradient
of the approximate function f.sub.5(x) around the corresponding
pixel (e.g., the pixel 3603 in FIG. 303) as the actual-world
features, thereby estimating the light signals in the actual world
1.
Then, for example, the image generating unit 103 shown in FIG. 289
predicts and generates an output image (FIG. 289) with higher
quality than the input image based upon the actual world features
detected by the actual world estimating means.
Specifically, at the image generating unit 103, for example, the
class classification adaptation processing unit 3501 shown in FIG.
289 predicts the pixel value of the pixel of interest (e.g., the
pixel of the predicted image shown in FIG. 289, and q' in the above
Expression (224)) based upon the pixel values of multiple pixels
around the pixel of interest in the input image in which a part of
continuity as the light signal in the actual world has been
lost.
On the other hand, for example, the class classification adaptation
processing correction unit 3502 shown in FIG. 289 predicts the
correction term (e.g., the pixel of the correction image
(subtraction predicted image) shown in FIG. 289, and u' in
Expression (227)) based upon the feature-amount image (actual world
estimation information) supplied from the actual world estimating
unit 102 shown in FIG. 289 for correcting the pixel value of the
pixel of interest of the predicted image predicted by the class
classification adaptation processing unit 3501.
Then, for example, the addition unit 3503 shown in FIG. 289
corrects the pixel value of the pixel of interest of the predicted
image predicted by the class classification adaptation processing
unit 3501 using the correction term predicted by the class
classification adaptation processing unit 3501 (e.g., computation
represented by Expression (224)).
Also, examples of components provided for the class classification
adaptation processing correction method include: the class
classification adaptation processing learning unit 3521 shown in
FIG. 291 for determining the prediction coefficients by learning,
stored in the coefficient memory 3514 shown in FIG. 290; and the
learning device 3504 shown in FIG. 291 including the class
classification adaptation processing correction learning unit 3561
shown in FIG. 291 for determining the correction coefficients by
learning, stored in the correction coefficient memory 3554 shown in
FIG. 299.
Specifically, for example, the class classification adaptation
processing learning unit 3521 shown in FIG. 292 includes: the
down-converter unit 3531 for performing down-converting processing
for the learning image data; the coefficient determining unit 3537
for generating the prediction coefficients by learning the relation
between the first tutor image and the first student image with the
learning image data as the first tutor image and with the learning
image data subjected to down-converting processing by the
down-converter unit 3531 as the first student image; and the region
extracting unit 3532 through the normal equation generating unit
3536.
The class classification adaptation processing learning unit 3521
further comprises a prediction computing unit 3538 for generating a
learning prediction image as image data for predicting a first
tutor image from a first student image, using a prediction
coefficient generated (determined) by the coefficient determining
unit 3537, for example.
On the other hand, for example, the class classification adaptation
processing correction learning unit 3561 shown in FIG. 300
includes: the data continuity detecting unit 3572 and the actual
world estimating unit 3573 for detecting the data continuity in the
first student image, detecting the actual-world features
corresponding to each pixel of the first student image based upon
the data continuity thus detected, and generating the
feature-amount image (specifically, the feature-amount image 3591
shown in FIG. 302, for example) with the value corresponding to the
detected actual-world feature as the pixel value, which is employed
as the second student image (e.g., the second student image in FIG.
300); the addition unit 3571 for generating the image data
(subtraction image) between the first student image and the
learning predicted image, which is used as the second tutor image;
the correction coefficient determining unit 3579 for generating the
correction coefficients by learning the relation between the second
tutor image and the second student image; and the region extracting
unit 3574 through the normal equation generating unit 3578.
Thus, the class classification adaptation processing correction
method enables output of an image more similar to the original
image (the signal in the actual world 1 which is to be input to the
sensor 2) as compared with other conventional methods including the
class classification adaptation processing.
Note that the difference between the class classification
adaptation processing and the simple interpolation processing is as
follows. That is to say, the class classification adaptation
processing enables reproduction of the components contained in the
HD image, which have been lost in the SD image, unlike the simple
interpolation. That is to say, as long as referring to only the
above Expressions (218) and (226), the class classification
adaptation processing looks like the same as the interpolation
processing using a so-called interpolation filter. However, with
the class classification adaptation processing, the prediction
coefficients d.sub.i and the correction coefficients g.sub.i
corresponding to the coefficients of the interpolation filter are
obtained by learning based upon the tutor data and the student data
(the first tutor image and the first student image, or the second
tutor image and the second student image), thereby reproducing the
components contained in the HD image. Accordingly, the class
classification adaptation processing described above can be said as
the processing having a function of improving the image quality
(improving the resolution).
While description has been made regarding an arrangement having a
function for improving the spatial resolution, the class
classification adaptation processing employs various kinds of
coefficients obtained by performing learning with suitable kinds of
the tutor data and the student data, thereby enabling various kinds
of processing for improving S/N (Signal to Noise Ratio), improving
blurring, and so forth.
That is to say, with the class classification adaptation
processing, the coefficients can be obtained with an image having a
high S/N as the tutor data and with the image having a reduced S/N
(or reduced resolution) generated based upon the tutor data as the
student data, for example, thereby improving S/N (or improving
blurring).
While description has been made regarding the image processing
device having a configuration shown in FIG. 3 s an arrangement
according to the present invention, an arrangement according to the
present invention is not restricted to the arrangement shown in
FIG. 3, rather, various modification may be made. That is to say,
an arrangement of the signal processing device 4 shown in FIG. 1 is
not restricted to the arrangement shown in FIG. 3, rather, various
modification may be made.
For example, the signal processing device having such a
configuration shown in FIG. 3 performs signal processing based upon
the data continuity contained in the signal in the actual world 1
serving as an image. Thus, the signal processing device having such
a configuration shown in FIG. 3 can perform signal processing with
high precision for the region where continuity is available for the
signal in the actual world 1, as compared with the signal
processing performed by other signal processing devices, thereby
outputting image data more similar to the signal in the actual
world 1, as a result.
However, the signal processing device having such a configuration
shown in FIG. 3 executes signal processing based upon continuity,
and accordingly, cannot execute signal processing with the same
precision for the region where clear continuity of the signal in
the actual world 1 is unavailable as processing for the region
where continuity is present, leading to output image data
containing an error as to the signal in the actual world 1.
Accordingly, an arrangement may be made further including another
device (or program) for performing signal processing which does not
employ continuity, in addition to the configuration of the signal
processing device shown in FIG. 3. With such an arrangement, the
signal processing device having the configuration shown in FIG. 3
executes signal processing for the region where continuity is
available for the signal in the actual world 1. On the other hand,
the additional device (or program or the like) executes the signal
processing for the region where clear continuity is unavailable for
the signal in the actual world 1. Note that such an arrangement
will be referred to as "hybrid method" hereafter.
Description will be made below with reference to FIG. 315 through
FIG. 328 regarding five specific hybrid method (which will be
referred to as "first hybrid method" through "fifth hybrid method"
hereafter).
Note that each function of the signal processing device employing
such a hybrid method may be realized by either of hardware and
software. That is to say, the block diagrams shown in FIG. 315
through FIG. 317, FIG. 321, FIG. 323, FIG. 325, and FIG. 327, may
be regarded to be either of hardware block diagrams or as software
block diagrams.
FIG. 315 shows a configuration example of a signal processing
device to which the first hybrid method is applied.
With the signal processing device shown in FIG. 315, upon reception
of the image data which an example of the data 3 (FIG. 1), image
processing as described later is performed based upon the input
image data (input image) so as to generate an image, and the
generated image (output image) is output. That is to say, FIG. 315
is a diagram which shows a configuration of the image processing
device 4 (FIG. 1) which is an image processing device.
The input image (image data which is an example of the data 3)
input to the image processing device 4 is supplied to a data
continuity detecting unit 4101, an actual world estimating unit
4102, and an image generating unit 4104.
The data continuity detecting unit 4101 detects the data continuity
from the input image, and supplies data continuity information
which indicates the detected continuity to the actual world
estimating unit 4102 and the image generating unit 4103.
As described above, the data continuity detecting unit 4101 has
basically the same configuration and functions as with the data
continuity detecting unit 101 shown in FIG. 3. Accordingly, the
data continuity detecting unit 4101 may have various kinds of
configurations described above.
Note that the data continuity detecting unit 4101 further has a
function for generating information for specifying the region of a
pixel of interest (which will be referred to as "region specifying
information" hereafter), and supplies the generated information to
a region detecting unit 4111.
The region specifying information used here is not restricted in
particular, rather, an arrangement may be made in which new
information is generated after the time that the data continuity
information has been generated, or an arrangement may be made in
which such information is generated as accompanying information of
the data continuity information at the same time.
Specifically, an estimation error may be employed as the region
specifying information, for example. That is to say, for example,
the estimation error is obtained as accompanying information at the
time of the data continuity detecting unit 4101 computing the angle
employed as the data continuity information using the least square
method. The estimation error may be employed as the region
specifying information.
The actual world estimating unit 4102 estimates the signal in the
actual world 1 (FIG. 1) based upon the input image and the data
continuity information supplied from the data continuity detecting
unit 4101. That is to say, the actual world estimating unit 4102
estimates the image which is the signal in the actual world 1, and
which is to be input to the sensor 2 (FIG. 1) in the stage where
the input image has been acquired. The actual world estimating unit
4102 supplies the actual world estimating information to the image
generating unit 4103 for indicating the estimation results of the
signal in the actual world 1.
As described above, the actual world estimating unit 4102 has
basically the same configuration and functions as with the actual
world estimating unit 102 shown in FIG. 3. Accordingly, the actual
world estimating unit 4102 may have various kinds of configurations
as described above.
The image generating unit 4103 generates a signal similar to the
signal in the actual world 1 based upon the actual world estimation
information indicating the estimated signal in the actual world 1
supplied from the actual world estimating unit 4102, and supplies
the generated signal to a selector 4112. Alternatively, the image
generating unit 4103 generates a signal closer to the signal of the
actual world 1 based upon: the data continuity information for
indicating the estimated signal in the actual world 1 supplied from
the data continuity detecting unit 4101; and the actual world
estimation information supplied from the actual world estimating
unit 4102, and supplies the generated signal to the selector
4112.
That is to say, the image generating unit 4103 generates an image
similar to the image of the actual world 1 based upon the actual
world estimation information, and supplies the generated image to
the selector 4112. Alternatively, the image generating unit 4103
generates an image more similar to the image of the actual world 1
based upon the data continuity information and the actual world
estimation information, and supplies the generated image to the
selector 4112.
As described above, the image generating unit 4103 has basically
the same configuration and functions as with the image generating
unit 103 shown in FIG. 3. Accordingly, the image generating unit
4103 may have various kinds of configurations as described
above.
The image generating unit 4104 performs predetermined image
processing for the input image so as to generate an image, and
supplies the generated image to the selector 4112.
Note that the image processing executed by the image generating
unit 4104 is not restricted in particular as long as employing the
image processing other than those employed in the data continuity
detecting unit 4101, the actual world estimating unit 4102, and the
image generating unit 4103.
For example, the image generating unit 4104 can perform
conventional class classification adaptation processing. FIG. 316
shows an configuration example of the image generating unit 4104
for executing the class classification adaptation processing. Note
that detailed description with reference to FIG. 316 will be made
later, i.e., detailed description will be made later regarding the
image generating unit 4104 for executing the class classification
processing. Also, description will be made later regarding the
class classification adaptation processing at the same time as with
description with reference to FIG. 316.
A continuity region detecting unit 4105 includes a region detecting
unit 4111 and a selector 4112.
The region detecting unit 4111 detects whether the image (pixel of
interest) supplied to the selector 4112 belongs to the continuity
region or non-continuity region based upon the region specifying
information supplied from the data continuity detecting unit 4101,
and supplies the detection results to the selector 4112.
Note that the region detection processing executed by the region
detecting unit 4111 is not restricted in particular. For example,
the aforementioned estimation error may be supplied as the region
specifying information. In this case, an arrangement may be made in
which in a case that the estimation error thus supplied is smaller
than a predetermined threshold, the region detecting unit 4111
determines that the pixel of interest of the input image belongs to
the continuity region, and in a case that the estimation error thus
supplied is greater than the predetermined threshold, determination
is made that the pixel of interest of the input image belongs to
the non-continuity region.
The selector 4112 selects one of the image supplied from the image
generating unit 4103 and the image supplied from the image
generating unit 4104 based upon the detection results supplied from
the region detecting unit 4111, and externally outputs the selected
image as an output image.
That is to say, in a case that the region detecting unit 4111 has
determined that the pixel of interest belongs to the continuity
region, the selector 4112 selects the image supplied from the image
generating unit 4103 (pixel corresponding to the pixel of interest
of the input image, generated by the image generating unit 4103) as
an output image.
On the other hand, in a case that the region detecting unit 4111
has determined that the pixel of interest belongs to the
non-continuity region, the selector 4112 selects the image supplied
from the image generating unit 4104 (pixel corresponding to the
pixel of interest of the input image, generated by the image
generating unit 4104) as an output image.
Note that the selector 4112 may output an output image in
increments of a pixel (i.e., may output an output image for each
selected pixel), or an arrangement may be made in which the pixels
subjected to the processing are stored until completion of the
processing for all the pixels, and all the pixels are output at the
same time (with the entire output image at once) when the
processing of all the pixels is completed.
Next, detailed description will be made regarding the image
generating unit 4104 for executing the class classification
adaptation processing which is an example of image processing with
reference to FIG. 316.
In FIG. 316, let us say that the class classification adaptation
processing executed by the image generating unit 4104 is processing
for improving the spatial resolution of an input image, for
example. That is to say, let us say that the class classification
adaptation processing is processing for converting an input image
with a standard resolution into a predicted image which is an image
with a high resolution.
Note that the image having a standard resolution will be referred
to as "SD (Standard Definition) image" hereafter as appropriate,
and the pixel making up the SD image will be referred to as "SD
pixel" as appropriate.
On the other hand, the image having a high resolution will be
referred to as "HD (High Definition) image" hereafter as
appropriate, and the pixel making up the HD image will be referred
to as "HD pixel" as appropriate.
Specifically, the class classification adaptation processing
executed by the image generating unit 4104 is as follows.
That is to say, in order to obtain the HD pixel of the predicted
image (HD image) corresponding to the pixel of interest (SD pixel)
of the input image (SD image), first, the features is obtained for
the SD pixels formed of the pixel of interest and the pixels
therearound (Such SD pixels will be also referred to as "class
taps" hereafter), and the class is identified for each class tap
based upon the features thereof by selecting one from the classes
prepared beforehand in association with the features (i.e., the
class code of the class-tap set is identified).
Then, product-sum is computed using: the coefficients of the one
selected from the multiple coefficient sets prepared beforehand
(each coefficient set corresponds to a certain class code) based
upon the identified class code; and the SD pixels formed of the
pixel of interest and the SD pixels therearound (Such SD pixels of
the input image will be also referred to as "prediction taps"
hereafter. Note that the prediction taps may match the class taps),
thereby obtaining the HD pixel of the predicted image (HD image)
corresponding to the pixel of interest (SD pixel) of the input
image (SD image).
More specifically, in FIG. 1, upon input of the signal in the
actual world 1 (light-intensity distribution) to the sensor 2, the
sensor 2 outputs an input image.
In FIG. 316, the input image (SD image) is supplied to region
extracting units 4121 and 4125 of the image generating unit 4104.
The region extracting unit 4125 extracts class taps (SD pixels
positioned at a predetermined region including the pixel of
interest (SD pixel)) necessary for class classification, from the
input image thus supplied, and outputs the extracted class taps to
a pattern detecting unit 4122. The pattern detecting unit 4122
detects the pattern of the input image based upon the class taps
thus input.
The class code determining unit 4123 determines the class code
based upon the pattern detected by the pattern detecting unit 4122,
and outputs the determined class code to coefficient memory 4124
and the region extracting unit 4125. The coefficient memory 4124
stores the coefficients for each class code obtained by learning.
The coefficient memory 4124 reads out the coefficients
corresponding to the class code input from the class code
determining unit 4123, and outputs the coefficients thus read, to a
prediction computing unit 4126.
Note that description will be made later regarding the learning
processing for obtaining the coefficients stored in the coefficient
memory 4124 with reference to the block diagram of the learning
device shown in FIG. 317.
Note that the coefficients stored in the coefficient memory 4124
are used for generating the predicted image (HD image) as described
later. Accordingly, the coefficients stored in the coefficient
memory 4124 will be referred to as "prediction coefficients"
hereafter.
The region extracting unit 4125 extracts the prediction taps (SD
pixels positioned at a predetermined region including the pixel of
interest) necessary for predicting and generating the predicted
image (HD image), from the input image (SD image) input from the
sensor 2 based upon the class code input from the class code
determining unit 4123 in response to the class code, and outputs
the extracted prediction taps to the prediction computing unit
4126.
The prediction computing unit 4126 executes product-sum computation
using the prediction taps input from the region extracting unit
4125 and the prediction coefficients input from the coefficient
memory 4124, thereby generating the HD pixel of the predicted image
(HD image) corresponding to the pixel of interest (SD pixel) of the
input image (SD image). Then, the prediction computing unit 4126
outputs the generated HD pixel to the selector 4112.
More specifically, the coefficient memory 4124 outputs the
prediction coefficients corresponding to the class code supplied
from the class code determining unit 4123 to the prediction
computing unit 4126. The prediction computing unit 4126 executes
product-sum computation represented by the following Expression
(240) using: the prediction taps extracted from the pixel value in
a predetermined pixel region of the input image supplied from the
region extracting unit 4125; and the prediction coefficients
supplied from the coefficient memory 4124, thereby obtaining (i.e.,
predicting and estimating) the HD pixel corresponding to the
predicted image (HD image).
'.times..times. ##EQU00147##
In Expression (240), q' represents the HD pixel of the predicted
image (HD image). Each of c.sub.i (i represents an integer of 1
through n) represents the corresponding prediction tap (SD pixel).
On the other hand, each of d.sub.i represents the corresponding
prediction coefficient.
As described above, the image generating unit 4104 predicts and
estimates the corresponding HD image based upon the SD image (input
image), and accordingly, in this case, the HD image output from the
image generating unit 4104 is referred to as a "predicted
image".
FIG. 317 shows a learning device (device for calculating the
prediction coefficients) for determining such prediction
coefficients (d.sub.i in Expression (237)) stored in the
coefficient memory 4124 of the image generating unit 4104.
In FIG. 317, a certain image is input to a down-converter unit 4141
and a normal equation generating unit 4146 as a tutor image (HD
image).
The down-converter unit 4146 generates a student image (SD image)
with a lower resolution than the input tutor image (HD image) based
upon the tutor image thus input (i.e., performs down-converting
processing for the tutor image, thereby obtaining a student image),
and outputs the generated student image to region extracting units
4142 and 4145.
As described above, a learning device 4131 includes the
down-converter unit 4141, and accordingly, there is no need to
prepare a higher-resolution image as the tutor image (HD image),
corresponding to the input image from the sensor 2 (FIG. 1). The
reason is that the student image (with a reduced resolution)
obtained by performing the down-converting processing for the tutor
image may be employed as an SD image. In this case, the tutor image
corresponding to the student image may be employed as an HD image.
Accordingly, the input image from the sensor 2 may be employed as
the tutor image without any conversion.
The region extracting unit 4142 extracts the class taps (SD pixels)
necessary for class classification, from the student image (SD
image) supplied from the down-converter unit 4141, and outputs the
extracted class taps to a pattern detecting unit 4143. The pattern
detecting unit 4143 detects the pattern of the class taps thus
input, and outputs the detection results to a class code
determining unit 4144. The class code determining unit 4144
determines the class code corresponding to the input pattern, and
outputs the determined class code to the region extracting unit
4145 and the normal equation generating unit 4146,
respectively.
The region extracting unit 4145 extracts the prediction taps (SD
pixels) from the student image (SD image) input from the
down-converter unit 4141, based upon the class code input from the
class code determining unit 4144, and outputs the extracted
prediction taps to the normal equation generating unit 4146.
Note that the aforementioned region extracting unit 4142, the
pattern detecting unit 4143, the class code determining unit 4144,
and the region extracting unit 4145, have basically the same
configurations and functions as with the region extracting unit
4121, the pattern detecting unit 4122, the class code determining
unit 4123, and the region extracting unit 4125, of the image
generating unit 4104 shown in FIG. 316, respectively.
The normal equation generating unit 4146 generates a normal
equation for each of all the class codes input from the class code
determining unit 4144 based upon the prediction taps (SD pixels) of
the student image (SD image) input from the region extracting unit
4145 and the HD pixels of the tutor image (HD image) for each class
code, and supplies the generated normal equation to a coefficient
determining unit 4147.
Upon reception of the normal equation corresponding to a certain
class code from the normal equation generating unit 4146, the
coefficient determining unit 4147 computes the prediction
coefficients using the normal equation, and stores the computed
prediction coefficients in the coefficient memory 4142 in
association with the class code.
Now, detailed description will be made regarding the normal
equation generating unit 4146 and the coefficient determining unit
4147.
In the above Expression (240), each of the prediction coefficients
d.sub.i is undetermined before learning. The learning processing is
performed by inputting the multiple HD pixels of the tutor image
(HD image) for each class code. Let us say that there are m HD
pixels corresponding to a certain class code. In this case, with
the m HD pixels as q.sub.k (k represents an integer of 1 through
m), the following Expression (241) is introduced from the
Expression (240).
.times..times. ##EQU00148##
That is to say, the Expression (241) indicates that a certain HD
pixel q.sub.k can be predicted and estimated by executing
computation represented by the right side thereof. Note that in
Expression (241), e.sub.k represents an error. That is to say, the
HD pixel q.sub.k' of the predicted image (HD image) obtained as
computation results by computing the right side does not exactly
match the actual HD pixel q.sub.k, but contains a certain error
e.sub.k.
With the present embodiment, the prediction coefficients d.sub.i
are obtained by learning processing such that the sum of squares of
the errors e.sub.k shown in Expression (241) exhibits the minimum,
thereby obtaining the optimum prediction coefficients d.sub.i for
predicting the actual HD pixel q.sub.k.
Specifically, with the present embodiment, the optimum prediction
coefficients d.sub.i are determined as a unique solution by
learning processing using the least square method based upon the m
HD pixels q.sub.k (wherein m is an integer greater than n)
collected by learning, for example.
That is to say, the normal equation for obtaining the prediction
coefficients d.sub.i in the right side of Expression (241) using
the least square method is represented by the following Expression
(242).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..function..-
times..times..times..times..times..times..times..times..times.
##EQU00149##
That is to say, with the present embodiment, the normal equation
represented by Expression (242) is generated and solved, thereby
determining the prediction coefficients d.sub.i as a unique
solution.
Specifically, with the component matrices forming the normal
equation represented by Expression (242) defined as the matrices
represented by Expressions (243) through (245), the normal equation
is represented by the following Expression (246).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times.
##EQU00150##
As can be understood from Expression (244), each component of the
matrix D.sub.MAT is the prediction coefficient d.sub.i which is to
be obtained. With the present embodiment, in the event that the
matrix C.sub.MAT, which is the left side of Expression (246), and
the matrix Q.sub.MAT, which is the right side thereof, are
determined, the matrix D.sub.MAT (i.e., prediction coefficient
d.sub.i) can be calculated with the matrix solution method.
More specifically, as can be understood from Expression (243), each
component of the matrix C.sub.MAT can be calculated as long as the
prediction taps c.sub.ik are known. The prediction taps c.sub.ik
are extracted by the region extracting unit 4145. With the present
embodiment, the normal equation generating unit 4146 can compute
each component of the matrix C.sub.MAT using the prediction tap
c.sub.ik supplied from the region extracting unit 4145.
On the other hand, as can be understood from Expression (245), each
component of the matrix Q.sub.MAT can be calculated as long as the
prediction taps c.sub.ik and the HD pixels q.sub.k are known. Note
that the prediction taps C.sub.ik are the same as those used in the
matrix C.sub.MAT, and the HD pixel q.sub.k is the HD pixel of the
tutor image corresponding to the pixel of interest (SD pixel of the
student image) included in the prediction taps c.sub.ik. With the
present embodiment, the normal equation generating unit 4146 can
compute each component of the matrix Q.sub.MAT using the prediction
taps c.sub.ik supplied from the region extracting unit 4145 and the
tutor image.
As described above, the normal equation generating unit 4146
computes each component of the matrix C.sub.MAT and each component
of the matrix Q.sub.MAT for each class code, and supplies the
computation results to the coefficient determining unit 4147 in
association with the class code.
The coefficient determining unit 4147 computes the prediction
coefficients d.sub.i each of which is the component of the matrix
D.sub.MAT represented by the above Expression (246) based upon the
normal equation corresponding to a certain class code supplied.
Specifically, the normal equation represented by the above
Expression (246) is transformed as represented by the following
Expression (247).
.times. ##EQU00151##
In Expression (247), each component of the matrix D.sub.MAT on the
left side thereof is the prediction coefficient d.sub.i which is to
be obtained. Note that each component of the matrix C.sub.MAT and
each component of the matrix Q.sub.MAT are supplied from the normal
equation generating unit 4146. With the present embodiment, upon
reception of each component of the matrix C.sub.MAT and each
component of the matrix Q.sub.MAT corresponding to a certain class
code from the normal equation generating unit 4146, the coefficient
determining unit 4147 computes matrix computation represented by
the right side of Expression (247) so as to calculate the Matrix
D.sub.MAT, and stores the computation results (prediction
coefficients d.sub.i) in the coefficient memory 4124 in association
with the class code.
Note that as described above, the difference between the class
classification adaptation processing and the simple interpolation
processing is as follows. That is to say, the class classification
adaptation processing enables reproduction of the component signals
contained in the HD image, which have been lost in the SD image,
unlike the simple interpolation, for example. That is to say, as
long as referring to only the above Expression (240), the class
classification adaptation processing looks like the same as the
interpolation processing using a so-called interpolation filter.
However, with the class classification adaptation processing, the
prediction coefficients d.sub.i corresponding to the coefficients
of the interpolation filter are obtained by learning based upon the
tutor data and the student data, thereby reproducing the components
contained in the HD image. Accordingly, the class classification
adaptation processing described above can be said as the processing
having a function of improving the image quality (improving the
resolution).
While description has been made regarding an arrangement having a
function for improving the spatial resolution, the class
classification adaptation processing employs various kinds of
coefficients obtained by performing learning with suitable kinds of
the tutor data and the student data, thereby enabling various kinds
of processing for improving S/N (Signal to Noise Ratio), improving
blurring, and so forth.
That is to say, with the class classification adaptation
processing, the coefficients can obtained with image data having a
high S/N as the tutor data and with the image having a reduced S/N
(or reduced resolution) generated based upon the tutor image as the
student image, for example, thereby improving S/N (or improving
blurring).
The above is description regarding the configurations of the image
generating unit 4104 and the learning device 4131 thereof for
executing the class classification adaptation processing.
Note that while the image generating unit 4104 may have a
configuration for executing image processing other than the class
classification adaptation processing as described above,
description will be made regarding the image generating unit 4104
having the same configuration as shown in FIG. 316 described above
for convenience of description. That is to say, let us say that the
image generating unit 4104 executes the class classification
adaptation processing so as to generate an image with higher
spatial resolution than the input image, and supplies the generated
image to the selector 4112.
Next, description will be made regarding signal processing
performed by the signal processing device (FIG. 315) employing the
first hybrid method with reference to FIG. 318.
Let us say that with the present embodiment, the data continuity
detecting unit 4101 computes angle (angle between: the continuity
direction (which is one spatial direction) around the pixel of
interest of the image, which represents the signal in the actual
world 1 (FIG. 1); and X-direction which is another spatial
direction (the direction parallel with a certain side of the
detecting element of the sensor 2), using the least square method,
and outputs the computed angle as data continuity information.
Also, the data continuity detecting unit 4101 outputs the
estimation error (error of the computation using the least square
method) calculated as accompanying computation results at the time
of computation of the angle, which is used as the region specifying
information.
In FIG. 1, upon input of the signal, which is an image, in the
actual world 1 to the sensor 2, the input image is output from the
sensor 2.
As shown in FIG. 315, the input image is input to the image
generating unit 4104, as well as to the data continuity detecting
unit 4101, and the actual world estimating unit 4102.
Then, in Step S4101 shown in FIG. 318, the image generating unit
4104 executes the aforementioned class classification adaptation
processing with a certain SD pixel of the input image (SD image) as
the pixel of interest, thereby generating the HD pixel (HD pixel
corresponding to the pixel of interest) of the predicted image (HD
image). Then, the image generating unit 4104 supplies the generated
HD pixel to the selector 4112.
Note that in order to distinguish between the pixel output from the
image generating unit 4104 and the pixel output from the image
generating unit 4103, the pixel output from the image generating
unit 4104 will be referred to as a "first pixel", and the pixel
output from the image generating unit 4103 will be referred to as a
"second pixel", hereafter.
Also, such processing executed by the image generating unit 4104
(the processing in Step S4101, in this case) will be referred to as
"execution of the class classification adaptation processing"
hereafter. Detailed description will be made later regarding an
example of the "execution of class classification adaptation
processing" with reference to the flowchart shown in FIG. 319.
On the other hand, in Step S4102, the data continuity detecting
unit 4101 detects the angle corresponding to the continuity
direction, and computes the estimation error thereof. The detected
angle is supplied to the actual world estimating unit 4102 and the
image generating unit 4103 as the data continuity information
respectively. On the other hand, the computed estimation error is
supplied to the region detecting unit 4111 as the region specifying
information.
In Step S4103, the actual world estimating unit 4102 estimates the
signal in the actual world 1 based upon the angle detected by the
data continuity detecting unit 4101 and the input image.
Note that the estimation processing executed by the actual world
estimating unit 4102 is not restricted in particular as described
above, rather, various kinds of techniques may be employed as
described above. Let us say that the actual world estimating unit
4102 approximates the function F (which will be referred to as
"light-signal function F" hereafter) which represents the signal in
the actual world 1, using a predetermined function f (which will be
referred to as "approximate function f" hereafter), thereby
estimating the signal (light-signal function F) in the actual world
1.
Also, let us say that the actual world estimating unit 4102
supplies the features (coefficients) of the approximate function f
to the image generating unit 4103 as the actual world estimation
information, for example.
In Step S4104, the image generating unit 4103 generates the second
pixel (HD pixel) based upon the signal in the actual world 1
estimated by the actual world estimating unit 4102, corresponding
to the first pixel (HD pixel) generated with the class
classification adaptation processing performed by the image
generating unit 4104, and supplies the generated second pixel to
the selector 4112.
With such a configuration, the features (coefficients) of the
approximate function f is supplied from the actual world estimating
unit 4102. Then, the image generating unit 4103 calculates the
integration of the approximate function f over a predetermined
integration range based upon the features of the approximate
function f thus supplied, thereby generating the second pixel (HD
pixel), for example.
Note that the integration range is determined so as to generate the
second pixel with the same size (same resolution) as with the first
pixel (HD pixel) output from the image generating unit 4104. That
is to say, the integration range is determined to be a range along
the spatial direction with the same width as that of the second
pixel which is to be generated.
Note that the order of steps according to the present invention is
not restricted to an arrangement shown in FIG. 318 in which the
"execution of class classification adaptation processing" in Step
S4101 and a series of processing in Step S4102 through Step S4104
are executed in that order, rather, an arrangement may be made in
which the series of processing in Step S4102 through Step S4104 is
executed prior to the "execution of class classification adaptation
processing" in Step S4101. Also, an arrangement may be made in
which the "execution of class classification adaptation processing"
in Step S4101 and a series of processing in Step S4102 through Step
S4104 are executed at the same time.
In Step S4105, the region detecting unit 4111 detects the region of
the second pixel (HD pixel) generated with the processing in Step
S4104 performed by the image generating unit 4103 based upon the
estimation error (region specifying information) computed with the
processing in Step S4102 performed by the data continuity detecting
unit 4101.
Here, the second pixel is an HD pixel corresponding to the SD pixel
of the input image, which has been used as the pixel of interest by
the data continuity detecting unit 4101. Accordingly, the type
(continuity region or non-continuity region) of the region is the
same between the pixel of interest (SD pixel of the input image)
and the second pixel (HD pixel).
Note that the region specifying information output from the data
continuity detecting unit 4101 is the estimation error calculated
at the time of calculation of the angle around the pixel of
interest using the least square method.
With such a configuration, the region detecting unit 4111 makes
comparison between the estimation error with regard to the pixel of
interest (SD pixel of the input image) supplied from the data
continuity detecting unit 4101 and a predetermined threshold. As a
result of comparison, in the event that the estimation error is
less than the threshold, the region detecting unit 4111 detects
that the second pixel belongs to the continuity region. On the
other hand, in the event that the estimation error is equal to or
greater than the threshold, the region detecting unit 4111 detects
that the second pixel belongs to the non-continuity region. Then,
the detection results are supplied to the selector 4112.
Upon reception of the detection results from the region detecting
unit 4111, the selector 4112 determines whether or not the detected
region belongs to the continuity region in Step S4106.
In Step S4106, in the event that determination has been made that
the detected region belongs to the continuity region, the selector
4112 externally outputs the second pixel supplied from the image
generating unit 4103 as an output image in Step S4107.
On the other hand, in Step S4106, in the event that determination
has been made that the detected region does not belong to the
continuity region (i.e., belongs to the non-continuity region), the
selector 4112 externally outputs the first pixel supplied from the
image generating unit 4104 as an output image in Step S4108.
Subsequently, in Step S4109, determination is made whether or not
the processing has been performed for all the pixels. In the event
that determination has been made that the processing has not been
performed for all the pixels, the processing returns to Step S4101.
That is to say, the processing in Step S4101 through S4109 is
repeated until completion of the processing for all the pixels.
On the other hand, in Step S4109, in the event that determination
has been made that the processing has been performed for all the
pixels, the processing ends.
As described above, with an arrangement shown in the flowchart in
FIG. 318, the output image selected from the first pixel and the
second pixel is output in increments as an output image of a pixel
each time that the first pixel (HD pixel) and the second pixel (HD
pixel) are generated.
However, as described above, the present invention is not
restricted to such an arrangement in which the output data is
output in increments of a pixel, rather, an arrangement may be made
in which the output data is output in the form of an image, i.e.,
the pixels forming the image are output at the same time as an
output image, each time that the processing has been made for all
the pixels. Note that with such an arrangement, each of Step S4107
and Step S4108 further includes additional processing for
temporarily storing the pixels (first pixels or second pixels) in
the selector 4112 instead of outputting the pixel each time that
the pixel is generated, and outputting all the pixels at the same
time after the processing in Step S4109.
Next, the details of the "processing for executing class
classification processing" which the image generating unit 4104 of
which the configuration is shown in FIG. 316 executes will be
described with reference to the flowchart in FIG. 319 (e.g.,
processing in step S4101 in FIG. 318 described above).
Upon an input image (SD image) being input to the image generating
unit 4104 from the sensor 2, in step S4121 the region extracting
unit 4121 and region extracting unit 4125 each input the input
image.
In step S4122, the region extracting unit 4121 extracts from the
input image a pixel of interest (SD pixel) and pixels (SD pixels)
at positions each at relative positions as to the pixel of interest
set beforehand (one or more positions), as a class tap, and
supplies this to the pattern detecting unit 4122.
In step S4123, the pattern detecting unit 4122 detects the pattern
of the supplied class tap, and supplies this to the class code
determining unit 4123.
In step S4124, the class code determining unit 4123 determines a
class code from multiple class codes set beforehand, which matches
the pattern of the class tap that has been supplied, and supplies
this to each of the coefficient memory 4124 and region extracting
unit 4125.
In step S4125, the coefficient memory 4124 reads out a prediction
coefficient (group) to be used, from multiple prediction
coefficients (groups) determined by learning processing beforehand,
based on the class code that has been supplied, and supplies this
to the prediction computing unit 4126.
Note that learning processing will be described later with
reference to the flowchart in FIG. 320.
In step S4126, the region extracting unit 4125 extracts, as a
prediction tap, from the input image corresponding to the class
code supplied thereto a pixel of interest (SD pixel) and pixels (SD
pixels) at positions each at relative positions as to the pixel of
interest set beforehand (One or more positions, being positions set
independently from the position of the class tap. However, may be
the same position as the class tap), and supplies this to the
prediction computing unit 4126.
In step S4127, the prediction computing unit 4126 computes the
prediction tap supplied from the region extracting unit 4125, using
the prediction coefficient supplied from the coefficient memory
4124, and generates a prediction image (first pixel) which is
externally (in the example in FIG. 315, the selector 4112)
output.
Specifically, the prediction computing unit 4126 takes each
prediction tap supplied from the region extracting unit 4125 as
c.sub.i (wherein i is an integer from 1 to n) and also each
prediction coefficient supplied from the coefficient memory 4124 as
d.sub.i, and computes the right side of the above-described
Expression (240) so as to calculate an HD pixel q' at the pixel of
interest (SD pixel), and externally outputs this as a predetermined
pixel (a first pixel) of the prediction image (HD image). After
this, the processing ends.
Next, the learning processing (processing for generating prediction
coefficients to be used by the image generating unit 4104 by
learning) which the learning device 4131 (FIG. 317) performs with
regard to the image generating unit 4104, will be described with
reference to the flowchart in FIG. 320.
In step S4141, each of the down converter unit 4141 and normal
equation generating unit 4146 inputs a predetermined image supplied
thereto as a tutor image (HD image).
In step S4142, the down converter unit 4141 performs down
conversion (reduction in resolution) of the input tutor image and
generates a student image (SD image), which is supplied to each of
the region extracting unit 4142 and region extracting unit
4145.
In step S4143, the region extracting unit 4142 extracts class taps
from the student image supplied thereto, and outputs to the patter
detecting unit 4143. Note that the processing in step S4143 is
basically the same processing as step S4122 (FIG. 319) described
above.
In step S4144, the pattern detecting unit 4143 detects patterns for
determining the class code form the class tap supplied thereto, and
supplies this to the class code determining unit 4144. Note that
the processing in step S4144 is basically the same processing as
step S4123 (FIG. 319) described above.
In step S4145, the class code determining unit 4144 determines the
class code based on the pattern of the class tap supplied thereto,
and supplies this to each of the region extracting unit 4145 and
the normal equation generating unit 4146. Note that the processing
in step S4145 is basically the same processing as step S4124 (FIG.
319) described above.
In step S4146, the region extracting unit 4145 extracts a
prediction tap from the student image corresponding to the class
code supplied thereto, and supplies this to the normal equation
generating unit 4146. Note that the processing in step S4146 is
basically the same processing as step S4126 (FIG. 319) described
above.
In step S4147, the normal equation generating unit 4146 generates a
normal equation expressed as the above-described Expression (242)
(i.e., Expression (243)) from the prediction tap (SD pixels)
supplied from the region extracting unit 4145 and a predetermined
HD pixel from the tutor image (HD image), and correlates the
generated normal equation with the class code supplied from the
class code determining unit 4144, and supplies this to the
coefficient determining unit 4147.
In step S4148, the coefficient determining unit 4147 solves the
supplied normal equation and determines the prediction coefficient,
i.e., calculates the prediction coefficient by computing the right
side of the above-described Expression (247), and stores this in
the coefficient memory 4124 in a manner correlated with the class
code supplied thereto.
Subsequently, in step S4149, determination is made regarding
whether or not processing has been performed for all pixels, and in
the event that determination is made that processing has not been
performed for all pixels, the processing returns to step S4143.
That is to say, the processing of steps S4143 through S4149 is
repeated until processing of all pixels ends.
Then, upon determination being made in step S4149 that processing
has been performed for all pixels, the processing ends.
Next, second third hybrid method will be described with reference
to FIG. 321 and FIG. 322.
FIG. 321 illustrates a configuration example of a signal processing
device to which the second hybrid method has been applied.
In FIG. 321, the portions which corresponding to the signal
processing device to which the first hybrid method has been applied
(FIG. 315) are denoted with corresponding symbols.
In the configuration example in FIG. 315 (the first hybrid method),
region identifying information is output from the data continuity
detecting unit 4101 and input to the region detecting unit 4111,
but with the configuration example shown in FIG. 321 (second hybrid
method), the region identifying information is output from the
actual world estimating unit 4102 and input to the region detecting
unit 4111.
This region identifying information is not restricted in
particular, and may be information newly generated following the
actual world estimating unit 4102 estimating signals of the actual
world 1 (FIG. 1), or may be information generated accessory to a
case of signals of the actual world 1 being estimated.
Specifically, for example, estimation error may be used as region
identifying information.
Now, description will be made regarding estimation error.
As described above, the estimated error output from the data
continuity detecting unit 4101 (region identifying information in
FIG. 315) is the estimation error calculated in an accessorial
manner while carrying out least-square computation in the event
that the continuity detecting information output from the data
continuity detecting unit 4101 is the angle, and the angle is
computed by the least-square method, for example.
Conversely, the estimation error (region identifying information in
FIG. 321) output from the actual world estimating unit 4102 is, for
example, mapping error.
That is to say, the actual world 1 signals are estimated by the
actual world estimating unit 4102, so pixels of an arbitrary
magnitude can be generated (pixel values can be calculated) from
the estimated actual world 1 signals. Here, in this way, generating
a new pixel is called mapping.
Accordingly, following estimating the actual world 1 signals, the
actual world estimating unit 4102 generates (maps) a new pixel from
the estimated actual world 1 signals, at the position where the
pixel of interest of the input image (the pixel used as the pixel
of interest in the case of the actual world 1 being estimated) was
situated. That is to say, the actual world estimating unit 4102
performs prediction computation of the pixel value of the pixel of
interest in the input image, from the estimated actual world 1
signals.
The actual world estimating unit 4102 then computes the difference
between the pixel value of the newly-mapped pixel (the pixel value
of the pixel of interest of the input image that has been
predicted) and the pixel value of the pixel of interest of the
actual input image. This difference is called mapping error.
By computing the mapping error (estimation error), the actual world
estimating unit 4102 can thus supply the computed mapping error
(estimation error) to the region detecting unit 4111 as region
identifying information.
While the processing for region detection which the region
detecting unit 4111 performs is not particularly restricted, as
described above, in the event of the actual world estimating unit
4102 supplying the above-described mapping error (estimation error)
to the region detecting unit 4111 as region identifying information
for example, the pixel of interest of the input image is detected
as being a continuity region in the event that the supplied mapping
error (estimation error) is smaller than a predetermined threshold
value, and on the other hand, the pixel of interest of the input
image is detected as being a non-continuity region in the event
that the supplied mapping error (estimation error) is equal to or
greater than a predetermined threshold value.
Other configurations are basically the same as shown in FIG. 315.
That is to say, the signal processing device to which the second
hybrid method is applied (FIG. 321) is also provided with the data
continuity detecting unit 4101, actual world estimating unit 4102,
image generating unit 4103, image generating unit 4104, and
continuity region detecting unit 4105 (region detecting unit 4111
and selector 4112), which have basically the same configurations
and functions as those of the signal processing device (FIG. 315)
to which the first hybrid method is applied.
FIG. 322 is a flowchart describing the processing of the signal
processing device of the configuration shown in FIG. 321 (signal
processing of the second hybrid method).
The signal processing of the second hybrid method is similar to the
signal processing of the first hybrid method (the processing shown
in the flowchart in FIG. 318). Accordingly, here, explanation of
processing described with regard to the first hybrid method will be
omitted as suitable, and description will proceed around the
processing according to the second hybrid method which differs from
the processing according to the first hybrid method with reference
to the flowchart in FIG. 322.
Note that here, as with the case of the first hybrid method, let us
say that the data continuity detecting unit 4101 uses the
least-square method to compute an angle (an angle between the
direction of continuity (spatial direction) at the pixel of
interest of the actual world 1 (FIG. 1) signals and the X direction
which is one direction in the spatial direction (a direction
parallel to a predetermined one side of the detecting elements of
the sensor 2 (FIG. 1)), and outputs the computed angle as data
continuity information.
However, while the data continuity detecting unit 4101 supplies the
region identifying information (e.g., estimated error) to the
region detecting unit 4111 in the first hybrid method as described
above, with the second hybrid method, the actual world estimating
unit 4102 supplies the region identifying information (e.g.,
estimation error (mapping error)) to the region detecting unit
4111.
Accordingly, with the second hybrid method, the processing of step
S4162 is executed as the processing of the data continuity
detecting unit 4101. This processing is equivalent to the
processing in step S4102 in FIG. 318, in the first hybrid method.
That is to say, the data continuity detecting unit 4101 detects an
angle corresponding to the direction of continuity, based on the
input image, and supplies the detected angle as data continuity
information to each of the actual world estimating unit 4102 and
image generating unit 4103.
Also, in the second hybrid method, the processing of step S4163 is
executed as the processing of the actual world estimating unit
4102. This processing is equivalent to the processing in step S4103
in FIG. 318, in the first hybrid method. That is to say, the actual
world estimating unit 4102 estimates the actual world 1 (FIG. 1)
signals based on the angle detected by the data continuity
detecting unit 4101 at the processing in step S4162, and computes
the estimated error of the estimated actual world 1 signals, i.e.,
mapping error, and supplies this as region identifying information
to the region detecting unit 4111.
Other processing is basically the same as the processing of the
first hybrid method (the corresponding processing of the processing
shown in the flowchart in FIG. 318), so description thereof will be
omitted.
Next, a third hybrid method will be described with reference to
FIG. 323 and FIG. 324.
FIG. 323 illustrates a configuration example of a signal processing
device to which the third hybrid method has been applied.
In FIG. 323, the portions which corresponding to the signal
processing device to which the first hybrid method has been applied
(FIG. 315) are denoted with corresponding symbols.
In the configuration example in FIG. 315 (the first hybrid method),
the continuity region detecting unit 4105 is disposed downstream
from the image generating unit 4103 and the image generating unit
4104, but with the configuration example shown in FIG. 323 (third
hybrid method), the continuity region detecting unit 4161
corresponding thereto is disposed downstream from a data continuity
detecting unit 4101 and upstream from an actual world estimating
unit 4102 and image generating unit 4104.
Due to such difference in the layout positions, there is somewhat
of a difference between the continuity region detecting unit 4105
in the first hybrid method and the continuity region detecting unit
4161 in the third hybrid method. The continuity detecting unit 4161
will be described mainly around this difference.
The continuity region detecting unit 4161 comprises a region
detecting unit 4171 and execution command generating unit 4172. Of
these, the region detecting unit 4171 has basically the same
configuration and functions as the region detecting unit 4111 (FIG.
315) of the continuity region detecting unit 4105. On the other
hand, the functions of the execution command generating unit 4172
are somewhat different to those of the selector 4112 (FIG. 315) of
the continuity region detecting unit 4105.
That is to say, as described above, the selector 4112 according to
the first hybrid technique selects one of an image from the image
generating unit 4103 and an image from the image generating unit
4104, based on the detection results form the region detecting unit
4111, and outputs the selected image as the output image. In this
way, the selector 4112 inputs an image from the image generating
unit 4103 and an image from the image generating unit 4104, in
addition to the detection results form the region detecting unit
4111, and outputs an output image.
On the other hand, the execution command generating unit 4172
according to the third hybrid method selects whether the image
generating unit 4103 or the image generating unit 4104 is to
execute processing for generating a new pixel at the pixel of
interest of the input image (the pixel which the data continuity
detecting unit 4101 has taken as the pixel of interest), based on
the detection results of the region detecting unit 4171.
That is to say, in the event that the region detecting unit 4171
supplies detection results to the execution command generating unit
4172 to the effects that the pixel of interest of the input image
is a continuity region, the execution command generating unit 4172
selects the image generating unit 4103, and supplies the actual
world estimating unit 4102 with a command to start the processing
(hereafter, such a command will be referred to as an execution
command). The actual world estimating unit 4102 then starts the
processing thereof, generates actual world estimation information,
and supplies this to the image generating unit 4103. The image
generating unit 4103 generates a new image based on the supplied
actual world estimation information (data continuity information
additionally supplied from the data continuity detecting unit 4101
as necessary), and externally outputs this as an output image.
Conversely, in the event that the region detecting unit 4171
supplies detection results to the execution command generating unit
4172 to the effects that the pixel of interest of the input image
is a non-continuity region, the execution command generating unit
4172 selects the image generating unit 4104, and supplies the image
generating unit 4104 with an execution command. The image
generating unit 4104 then starts the processing, subjects the input
image to predetermined image processing (class classification
adaptation processing in this case), generates a new image, and
externally outputs this as an output image.
Thus, the execution command generating unit 4172 according to the
third hybrid method inputs the detection results to the region
detecting unit 4171 and outputs execution commands. That is to say,
the execution command generating unit 4172 does not input or output
images.
Note that the configuration other than the continuity region
detecting unit 4161 is basically the same as that in FIG. 315. That
is to say, the signal processing device to which the second hybrid
method is applied (the signal processing device in FIG. 323) also
is provided with the data continuity detecting unit 4101, actual
world estimating unit 4102, image generating unit 4103, and image
generating unit 4104, having basically the same configurations and
functions as the signal processing device to which the first hybrid
method is applied (FIG. 315).
However, with the third hybrid method, the actual world estimating
unit 4102 and the image generating unit 4104 do not each execute
the processing thereof unless an execution command is input from
the execution command generating unit 4172.
Now, with the example shown in FIG. 323, the output unit of the
image is in units of pixels. Accordingly, though not shown, an
image synthesizing unit may be further provided downstream of the
image generating unit 4103 and image generating unit 4104 for
example, in order to make the output unit to be the entire image of
one frame (in order to output all pixels at once).
This image synthesizing unit adds (synthesizes) the pixel values
output from the image generating unit 4103 and the image generating
unit 4104, and takes the added value as the pixel value of the
corresponding pixel. In this case, the one of the image generating
unit 4103 and the image generating unit 4104 which has not been
supplied with an execution command does not execute the processing
thereof, and constantly supplies a predetermined constant value
(e.g., 0) to the image synthesizing unit.
The image synthesizing unit repeatedly executes such processing for
all pixels, and upon ending processing for all pixels, externally
outputs all pixels at once (as one frame of image data).
Next, the signal processing of the signal processing device to
which the third hybrid method has been applied (FIG. 323) will be
described with reference to the flowchart in FIG. 324.
Note that here, as with the case of the first hybrid method, let us
say that the data continuity detecting unit 4101 uses the
least-square method to compute an angle (an angle between the
direction of continuity (spatial direction) at the position of
interest of the actual world 1 (FIG. 1) signals and the X direction
which is one direction in the spatial direction (a direction
parallel to a predetermined one side of the detecting elements of
the sensor 2 (FIG. 1)), and outputs the computed angle as data
continuity information.
Let us also say that the data continuity detecting unit 4101
outputs the estimated error calculated (error of least-square)
along with calculation of the angle as the region identifying
information.
In FIG. 1, upon the signals of the actual world 1 being cast into
the sensor 2, the sensor 2 outputs an input image.
In FIG. 323, this input image is input to the image generating unit
4104, and is also input to the data continuity detecting unit 4101
and the actual world estimating unit 4102.
Now, in step S4181 in FIG. 324, the data continuity detecting unit
4101 detects the angle corresponding to the direction of the
continuity based on the input image, and also computes the
estimated error thereof. The detected angle is supplied to is
supplied to each of the actual world estimating unit 4102 and the
image generating unit 4103, as data continuity information. Also,
the computed estimated error is supplied to the region detecting
unit 4171 as region identifying information.
Note that the processing of step S4181 is basically the same as the
processing of step S4102 (FIG. 318) described above.
Also, as described above, at this point (unless an execution
command is supplied from the execution command generating unit
4172), neither the actual world estimating unit 4102 nor the image
generating unit 4103 execute the processing thereof.
In step S4182, the region detecting unit 4171 detects the region of
the pixel of interest (the pixel to be taken as the pixel of
interest in the case of the data continuity detecting unit 4101
detecting the angle) in the input image, based on the estimated
error computed by the data continuity detecting unit 4101 (the
supplied region identifying information), and supplies the
detection results thereof to the execution command generating unit
4172. Note that the processing in step S4182 is basically the same
as the processing of step S4105 (FIG. 318) described above.
Upon the detection results of the region detecting unit 4171 being
supplied to the execution command generating unit 4172, in step
S4183 the execution command generating unit 4172 determines whether
or not the detected region is a continuity region. Note that the
processing of step S4183 is basically the same as the processing of
step S4106 (FIG. 318) described above.
In step S4183, in the event that determination is made that the
detected region is not a continuity region, the execution command
generating unit 4172 supplies an execution command to the image
generating unit 4104. the image generating unit 4104 then executes
"processing for executing class classification adaptation
processing" in step S4184, generates a first pixel (HD pixel at the
pixel of interest (SD pixel of the input image)), and in step S4185
externally outputs the first pixel generated by the class
classification adaptation processing, as an output image.
Note that the processing of step S4184 is basically the same as the
processing of step S4101 (FIG. 318) described above. That is to
say, the flowchart in FIG. 319 is a flowchart for describing the
details of processing in step S4184.
Conversely, in step S4183, in the event that determination is made
that the detected region is a continuity region, the execution
command generating unit 4172 supplies an execution command to the
actual world estimating unit 4102. In step S4186, the actual world
estimating unit 4102 then estimates the actual world 1 signals
based on the angle detected by the data continuity detecting unit
4101 and the input image. Note that the processing of step S4186 is
basically the same as the processing of step S4103 (FIG. 318)
described above.
In step S4187, the image generating unit 4103 generates a second
pixel (HD pixel) in the detected region (i.e., the pixel of
interest (SD pixel) in the input image), based on the actual world
1 signals estimated by the actual world estimating unit 4102, and
outputs the second pixel as an output image in step S4188. Note
that the processing of step S4187 is basically the same as the
processing of step S4104 (FIG. 318) described above.
Upon a first pixel or a second pixel being output as an output
image (following processing of step S4185 or step S4188), in step
S4189 determination is made regarding whether or not processing has
ended for all pixels, and in the event that processing of all
pixels has not ended yet, the processing returns to step S4181.
That is to say, the processing of steps S4181 through S4189 is
repeated until the processing of all pixels is ended.
Then, in step S4189, in the event that determination is made that
processing of all pixels has ended, the processing ends.
In this way, in the example of the flowchart in FIG. 324, each time
a first pixel (HD pixel) and second pixel (HD pixel) are generated,
the first pixel or second pixel are output in pixel increment as an
output image.
However, as described above, an arrangement wherein an image
synthesizing unit (not shown) is further provided at the furthest
downstream portion of the signal processing device having the
configuration shown in FIG. 323 (downstream of the image generating
unit 4103 and the image generating unit 4104) enables all pixels to
be output at once as an output image following processing of all
pixels having ended. In this case, the pixel (first pixel or second
pixel) is output not externally but to the image synthesizing unit
in the processing of step S4185 and step S4188. Then, before the
processing in step S4189, processing is added wherein the image
synthesizing unit synthesizes the pixel values of the pixels
supplied from the image generating unit 4103 and the pixel values
of the pixels supplied from the image generating unit 4104, and
following the processing of step S4189 for generating pixels of the
output image, processing is added wherein the image synthesizing
unit outputs all pixels.
Next, a fourth hybrid method will be described with reference to
FIG. 325 and FIG. 326.
FIG. 325 illustrates a configuration example of a signal processing
device to which the fourth hybrid method has been applied.
In FIG. 325, the portions which corresponding to the signal
processing device to which the third hybrid method has been applied
(FIG. 323) are denoted with corresponding symbols.
In the configuration example in FIG. 323 (the third hybrid method),
the region identifying information is input from the data
continuity detecting unit 4101 to the region detecting unit 4171,
but with the configuration example shown in FIG. 325 (fourth hybrid
method), region identifying information is output from the actual
world estimating unit 4102 and input to the region detecting unit
4171.
Other configurations are basically the same as that in FIG. 323.
That is to say, the signal processing device to which the fourth
hybrid method is applied (the signal processing device in FIG. 325)
also is provided with the data continuity detecting unit 4101,
actual world estimating unit 4102, image generating unit 4103,
image generating unit 4104, and continuity region detecting unit
4161 (region detecting unit 4171 and execution command generating
unit 4172) having basically the same configurations and functions
as the signal processing device to which the third hybrid method is
applied (FIG. 323).
Also, as with the third hybrid method, an arrangement may be made
wherein an image synthesizing unit is disposed downstream from the
image generating unit 4103 and image generating unit 4104, for
example, to output all pixels at once, though not shown in the
drawings.
FIG. 326 is a flowchart for describing the signal processing of the
signal processing device of the configuration shown in FIG. 325
(signal processing according to the fourth hybrid method).
The signal processing according to the fourth hybrid method is
similar to the signal processing according to the third hybrid
method (the processing shown in the flowchart in FIG. 324).
Accordingly, here, explanation of processing described with regard
to the third hybrid method will be omitted as suitable, and
description will proceed around the processing according to the
fourth hybrid method which differs from the processing according to
the third hybrid method, with reference to the flowchart in FIG.
326.
Note that here, as with the case of the third hybrid method, let us
say that the data continuity detecting unit 4101 uses the
least-square method to compute an angle (an angle between the
direction of continuity (spatial direction) at the pixel of
interest of the actual world 1 (FIG. 1) signals and the X direction
which is one direction in the spatial direction (a direction
parallel to a predetermined one side of the detecting elements of
the sensor 2 (FIG. 1)), and outputs the computed angle as data
continuity information.
However, while the data continuity detecting unit 4101 supplies the
region identifying information (e.g., estimated error) to the
region detecting unit 4171 in the third hybrid method as described
above, with the fourth hybrid method, the actual world estimating
unit 4102 supplies the region identifying information (e.g.,
estimation error (mapping error)) to the region detecting unit
4171.
Accordingly, with the fourth hybrid method, the processing of step
S4201 is executed as the processing of the data continuity
detecting unit 4101. This processing is equivalent to the
processing in step S4181 in FIG. 324, in the third hybrid method.
That is to say, the data continuity detecting unit 4101 detects an
angle corresponding to the direction of continuity, based on the
input image, and supplies the detected angle as data continuity
information to each of the actual world estimating unit 4102 and
image generating unit 4103.
Also, in the fourth hybrid method, the processing of step S4202 is
executed as the processing of the actual world estimating unit 4102
in step S4202. This processing is equivalent to the processing in
step S4182 in FIG. 318, in the third hybrid method. That is to say,
the actual world estimating unit 4102 estimates the actual world 1
(FIG. 1) signals based on the angle detected by the data continuity
detecting unit 4101, and computes the estimated error of the
estimated actual world 1 signals, i.e., mapping error, and supplies
this as region identifying information to the region detecting unit
4171.
Other processing is basically the same as the processing of the
third hybrid method (the corresponding processing of the processing
shown in FIG. 324), so description thereof will be omitted.
Next, a fifth hybrid method will be described with reference to
FIG. 327 and FIG. 328.
FIG. 327 illustrates a configuration example of a signal processing
device to which the fifth hybrid method has been applied.
In FIG. 327, the portions which corresponding to the signal
processing devices to which the third and fourth hybrid methods
have been applied (FIG. 323 and FIG. 325) are denoted with
corresponding symbols.
In the configuration example shown in FIG. 323 (third hybrid
method), one continuity region detecting unit 4161 is disposed
downstream of the data continuity detecting unit 4101 and upstream
of the actual world estimating unit 4102 and image generating unit
4104.
Also, in the configuration example shown in FIG. 325 (fourth hybrid
method), one continuity region detecting unit 4161 is disposed
downstream of the actual world estimating unit 4102 and upstream of
the image generating unit 4103 and image generating unit 4104.
Conversely, with the configuration example shown in FIG. 327 (fifth
hybrid method), the continuity region detecting until 4181 is
disposed downstream form the data continuity detecting unit 4101
but upstream from the actual world estimating unit 4102 and the
image generating unit 4101, as with the third hybrid method.
Further, as with the fourth hybrid method, a continuity region
detecting unit 4182 is disposed downstream from the actual world
estimating unit 4102 but upstream from the image generating unit
4103 and the image generating unit 4104.
The continuity region detecting unit 4181 and continuity region
detecting unit 4182 both basically have basically the same
configurations and functions as the continuity region detecting
unit 4161 (FIG. 323 or FIG. 325). That is to say, both the region
detecting unit 4191 and region detecting unit 4201 have basically
the same configuration and functions as the region detecting unit
4171.
Restated, the fifth hybrid method is a combination of the third
hybrid method and the fourth hybrid method.
That is to say, with the third hybrid method and the fourth hybrid
method, whether the pixel of interest of an input image is a
continuity region or a non-continuity region is determined based on
one region identifying information (in the case of the third hybrid
method, the region identifying information from the data continuity
detecting unit 4101, and in the case of the fourth hybrid method,
the region identifying information from the actual world estimating
unit 4102). Accordingly, the third hybrid method and the fourth
hybrid method could detect a region to be a continuity region even
though it should be a non-continuity region.
Accordingly, with the fifth hybrid method, following detection of
whether the pixel of interest of an input image is a continuity
region or a non-continuity region, based on region identifying
information from the data continuity detecting unit 4101 (this will
be called first region identifying information in the description
of the fifth hybrid method), further detection is made regarding
whether the pixel of interest of an input image is a continuity
region or a non-continuity region, based on region identifying
information from the actual world estimating unit 4102 (this will
be called second region identifying information in the description
of the fifth hybrid method).
In this way, with the fifth hybrid method, processing for region
detection is performed twice, so precision of detection of the
continuity region improves over that of the third hybrid method and
the fourth hybrid method. Further, with the first hybrid method and
the second hybrid method as well, only one continuity region
detecting unit 4105 (FIG. 315 or FIG. 321) is provided, as with the
case of the third hybrid method and the fourth hybrid method.
Accordingly, the detection precision of the continuity region
improves in comparison with the first hybrid method and the second
hybrid method as well. Consequently, output of image data closer to
signals of the actual world 1 (FIG. 1) than any of the first
through fourth hybrid methods can be realized.
However, it remains unchanged that even the first through fourth
hybrid methods use both the image generating unit 4104 which
performs conventional image processing, and devices or programs and
the like for generating image using data continuity, to which the
present invention is applied (i.e., the data continuity detecting
unit 4101, actual world estimating unit 4102, and image generating
unit 4103).
Accordingly, the first through fourth hybrid methods are capable of
outputting image data closer to signals of the actual world 1 (FIG.
1) than any of conventional signal processing devices or the signal
processing according to the present invention with the
configuration shown in FIG. 3.
On the other hand, from the perspective of processing speed, region
detection processing is required only once with the first through
fourth hybrid methods, and accordingly these are superior to the
fifth hybrid methods which performs region detection processing
twice.
Accordingly, the user (or manufacture) or the like can selectively
use a hybrid method which meets the quality of the output image
required, and the required processing time (the time until the
output image is output).
Note that other configurations in FIG. 327 are basically the same
as those in FIG. 323 or FIG. 325. That is to say, the signal
processing device to which the fifth hybrid method has been applied
(FIG. 327) is provided with the data continuity detecting unit
4101, actual world estimating unit 4102, image generating unit
4103, and image generating unit 4104, having basically the same
configurations and functions as with the signal processing devices
to which the third or fourth hybrid methods have been applied (FIG.
323 or FIG. 325).
However, with the fifth hybrid method, the actual world estimating
unit 4102 does not execute the processing thereof unless an
execution command is input from the execution command generating
unit 4192, the image generating unit 4103 does not unless an
execution command is input from the execution command generating
unit 4202, and the image generating unit 4104 does not unless an
execution command is input from the execution command generating
unit 4192 or the execution command generating unit 4202.
Also, in the fifth hybrid method as well, as with the third or
fourth hybrid methods, an arrangement may be made wherein an image
synthesizing unit is disposed downstream from the image generating
unit 4103 and image generating unit 4104 to output all pixels at
once, for example, though not shown in the drawings.
Next, the signal processing of the signal processing device to
which the fifth hybrid method has been applied (FIG. 327) will be
described with reference to the flowchart in FIG. 328.
Note that here, as with the case of the third and fourth hybrid
methods, let us say that the data continuity detecting unit 4101
uses the least-square method to compute an angle (an angle between
the direction of continuity (spatial direction) at the position of
interest of the actual world 1 (FIG. 1) signals and the X direction
which is one direction in the spatial direction (a direction
parallel to a predetermined one side of the detecting elements of
the sensor 2 (FIG. 1)), and outputs the computed angle as data
continuity information.
Let us also say here that the data continuity detecting unit 4101
outputs the estimated error calculated (error of least-square)
along with calculation of the angle as first region identifying
information, as with the case of the third hybrid method.
Let us further say that the actual world estimating unit 4102
outputs mapping error (estimation error) as second region
identifying information, as with the case of the fourth hybrid
method.
In FIG. 1, upon the signals of the actual world 1 being cast into
the sensor 2, the sensor 2 outputs an input image.
In FIG. 327, this input image is input to the image generating unit
4104, and is also input to the data continuity detecting unit 4101,
actual world estimating unit 4102, image generating unit 4103, and
image generating unit 4104.
Now, in step S4221 in FIG. 328, the data continuity detecting unit
4101 detects the angle corresponding to the direction of the
continuity based on the input image, and also computes the
estimated error thereof. The detected angle is supplied to is
supplied to each of the actual world estimating unit 4102 and the
image generating unit 4103, as data continuity information. Also,
the computed estimated error is supplied to the region detecting
unit 4191 as first region identifying information.
Note that the processing of step S4221 is basically the same as the
processing of step S4181 (FIG. 324) described above.
Also, as described above, at the current point, unless an execution
command is supplied from the execution command generating unit
4192), neither the actual world estimating unit 4102 nor the image
generating unit 4104 perform the processing thereof.
In step S4222, the region detecting unit 4191 detects the region of
the pixel of interest (the pixel to be taken as the pixel of
interest in the case of the data continuity detecting unit 4101
detecting the angle) in the input image, based on the estimated
error computed by the data continuity detecting unit 4101 (the
supplied first region identifying information), and supplies the
detection results thereof to the execution command generating unit
4192. Note that the processing in step S4222 is basically the same
as the processing of step S4182 (FIG. 324) described above.
Upon the detection results of the region detecting unit 4181 being
supplied to the execution command generating unit 4192, in step
S4223 the execution command generating unit 4192 determines whether
or not the detected region is a continuity region. Note that the
processing of step S4223 is basically the same as the processing of
step S4183 (FIG. 324) described above.
In step S4223, in the event that determination is made that the
detected region is not a continuity region (is a non-continuity
region), the execution command generating unit 4192 supplies an
execution command to the image generating unit 4104. The image
generating unit 4104 then executes "processing for executing class
classification adaptation processing" in step S4224, generates a
first pixel (HD pixel at the pixel of interest (SD pixel of the
input image)), and in step S4225 externally outputs the first pixel
generated by the class classification adaptation processing, as an
output image.
Note that the processing of step S4224 is basically the same as the
processing of step S4184 (FIG. 324) described above. That is to
say, the flowchart in FIG. 319 is also a flowchart for describing
the details of processing in step S4186. Also, the processing of
step S4225 is basically the same as the processing of step S4185
(FIG. 324) described above.
Conversely, in step S4223, in the event that determination is made
that the detected region is a continuity region, the execution
command generating unit 4192 supplies an execution command to the
actual world estimating unit 4102. In step S4226, the actual world
estimating unit 4102 then estimates the actual world 1 signals
based on the angle detected by the data continuity detecting unit
4101 and the input image in the processing of step S4221, and also
computes the estimation error (mapping error) thereof. The
estimated actual world 1 signals are supplied to the image
generating unit 4103 as actual world estimation information. Also,
the computed estimation error is supplied to the region detecting
unit 4201 as second region identifying information.
Note that the processing of step S4226 is basically the same as the
processing of step S4202 (FIG. 326) described above.
Also, as described above, at this point (unless an execution
command is supplied from the execution command generating unit 4192
or the execution command generating unit 4202), neither the image
generating unit 4103 nor the image generating unit 4104 execute the
processing thereof.
In step S4227, the region detecting unit 4201 detects the region of
the pixel of interest (the pixel to be taken as the pixel of
interest in the case of the data continuity detecting unit 4101
detecting the angle) in the input image, based on the estimated
error computed by the data continuity detecting unit 4101 (the
supplied second region identifying information), and supplies the
detection results thereof to the execution command generating unit
4202. Note that the processing in step S4227 is basically the same
as the processing of step S4203 (FIG. 326) described above.
Upon the detection results of the region detecting unit 4201 being
supplied to the execution command generating unit 4202, in step
S4228 the execution command generating unit 4202 determines whether
or not the detected region is a continuity region. Note that the
processing of step S4228 is basically the same as the processing of
step S4204 (FIG. 326) described above.
In step S4228, in the event that determination is made that the
detected region is not a continuity region (is a non-continuity
region), the execution command generating unit 4202 supplies an
execution command to the image generating unit 4104. The image
generating unit 4104 then executes "processing for executing class
classification adaptation processing" in step S4224, generates a
first pixel (HD pixel at the pixel of interest (SD pixel of the
input image)), and in step S4225 externally outputs the first pixel
generated by the class classification adaptation processing, as an
output image.
Note that the processing of step S4224 here is basically the same
as the processing of step S4205 (FIG. 326) described above. Also,
the processing of step S4225 here is basically the same as the
processing of step S4206 (FIG. 326) described above.
Conversely, in step S4228, in the event that determination is made
that the detected region is a continuity region, the execution
command generating unit 4202 supplies an execution command to the
image generating unit 4103. In step S4229, the image generating
unit 4103 then generates a second pixel (HD pixel) in the region
detected by the region detecting unit 4201 (i.e., the pixel of
interest (SD pixel) in the input image), based on the actual world
1 signals estimated by the actual world estimating unit 4102 (and
data continuity signals from the data continuity detecting unit
4101 as necessary). Then, in step S4230, the image generating unit
4103 externally outputs the generated second pixel as an output
image.
Note that the processing of steps S4229 and S4230 is each basically
the same as the processing of each of steps S4207 and S4208 (FIG.
326) described above.
Upon a first pixel or a second pixel being output as an output
image (following processing of step S4225 or step S4230), in step
S4231 determination is made regarding whether or not processing has
ended for all pixels, and in the event that processing of all
pixels has not ended yet, the processing returns to step S4221.
That is to say, the processing of steps S4221 through S4231 is
repeated until the processing of all pixels is ended.
Then, in step S4231, in the event that determination is made that
processing of all pixels has ended, the processing ends.
The hybrid method has been described so far as an example of an
embodiment of the signal processing device 4 (FIG. 1) according to
the present invention, with reference to FIG. 315 through FIG.
328.
As described above, with the hybrid method, another device (or
program or the like) which performs signal processing without using
continuity is further added to the signal processing device
according to the present invention having the configuration shown
in FIG. 3.
In other words, with the hybrid method, the signal processing
device (or program or the like) according to the present invention
having the configuration shown in FIG. 3 is added to a conventional
signal processing device (or program or the like).
That is to say, with the hybrid method, the continuity region
detecting unit 4105 shown in FIG. 315 or FIG. 321 for example,
detects data regions having data continuity of image data (e.g.,
the continuity region described in step S4106 in FIG. 318 or step
S4166 in FIG. 322) within image data wherein light signals of the
actual world 1 have been projected and a part of the continuity of
the light signals of the actual world 1 has been lost (e.g., the
input image in FIG. 315 or FIG. 321).
Also, the actual world estimating unit 4102 shown in FIG. 315 and
FIG. 321 estimates the light signals by estimating the lost
continuity of the light signals of the actual world 1, based on the
data continuity of the image data of which a part of the continuity
of the light signals of the actual world 1 has been lost.
Further, the data continuity detecting unit 4101 shown in FIG. 315
and FIG. 321 detects the angle of the data continuity of the image
data as to a reference axis (for example, the angle described in
step S4102 in FIG. 318 and step S4162 in FIG. 322), within image
data wherein light signals of the actual world 1 have been
projected and a part of the continuity of the light signals of the
actual world 1 has been lost. In this case, for example, the
continuity region detecting unit 4105 shown in FIG. 315 and FIG.
321 detects regions in the image data having data continuity based
on the angle, and the actual world estimating unit 4102 estimates
the light signals by estimating the continuity of the light signals
of the actual world 1 that has been lost, with regard to that
region.
However, in FIG. 315, the continuity region detecting unit 4105
detects regions of the input image having data continuity based on
the error between a model having continuity following the angle,
and the input image (that is, estimation error which is the region
identifying information in the drawing, computed by the processing
in step S4102 of FIG. 318).
Conversely, in FIG. 321, the continuity region detecting unit 4105
is disposed downstream from the actual world estimating unit 4102,
and selectively outputs (e.g., the selector 4112 in FIG. 321
executes the processing of steps S4166 through S4168 in FIG. 322)
an actual world model estimated by the actual world estimating unit
4102, based on error between an actual world model representing
light signals of the actual world 1 corresponding to the input
image computed by the actual world estimating unit 4102 and the
input image (i.e., estimation error (mapping error) of actual world
signals computed by the processing in step S4163 in FIG. 318, which
is region identifying information in the drawing, for example),
i.e., outputs an image output from the image generating unit
4103.
While the above description has been made with the example of FIG.
315 and FIG. 321, the same is true for FIG. 323, FIG. 325, and FIG.
327.
Accordingly, with the hybrid method, a device (or program or the
like) corresponding to the signal processing device of the
configuration shown in FIG. 3 executes signal processing for
portions of the actual world 1 signals where continuity exists
(regions of the image data having data continuity), and a
conventional signal processing device (or program or the like) can
execute signal processing for portions of the actual world 1
signals where there is no clear continuity. As a result, output of
image data closer to signals of the actual world (FIG. 1) than
either of conventional signal processing devices and the signal
processing according to the present invention of the configuration
shown in FIG. 3 can be realized.
Next, an example of directly generating an image from the data
continuity detecting unit 101 will be described with reference to
FIG. 329 and FIG. 330.
The data continuity detecting unit 101 shown in FIG. 329 is the
data continuity detecting unit 101 shown in FIG. 165 with an image
generating unit 4501 added thereto. The image generating unit 4501
acquires as actual world estimation information a coefficient of
the actual world approximation function f(x) output from the actual
world estimating unit 802, and generates and outputs an image by
reintegration of each pixel based on this coefficient.
Next, the data continuity detection processing in FIG. 329 will be
described with reference to the flowchart shown in FIG. 330. Note
that the processing in steps S4501 through S4504 and steps S4506
through S4511 of the flowchart in FIG. 330 is the same as the
processing in steps S801 through S810 in FIG. 166, so description
thereof will be omitted.
In step S4504, the image generating unit 4501 reintegrates each of
the pixels based on the coefficient input form the actual world
estimating unit 802, and generates and outputs an image.
Due to the above processing, the data continuity detecting unit 101
can output not only region information built also an image used for
the region determination (made up of pixels generated based on the
actual world estimation information).
Thus, with the data continuity detecting unit 101 shown in FIG.
329, the image generating unit 4501 is provided. That is to say,
the data continuity detecting unit 101 in FIG. 329 can generate
output images based on the data continuity of the input image.
Accordingly, a device having the configuration shown in FIG. 329
can be interpreted to be another embodiment of the signal
processing device (image processing device) 4 shown in FIG. 1,
rather than being interpreted as an embodiment of the data
continuity detecting unit 101.
Further, with the signal processing device to which the
above-described hybrid method is applied, a device having the
configuration shown in FIG. 329 (i.e., a signal processing device
having the same functions and configuration as the data continuity
detecting unit 101 in FIG. 329) can be applied as the signal
processing unit for subjecting the portions of the signals of the
actual world 1 where continuity exists, to signal processing.
Specifically, for example, with the signal processing device shown
in FIG. 315 to which the first hybrid method is applied, the signal
processing unit for subjecting the portions of the signals of the
actual world 1 where continuity exists, to signal processing, is
the data continuity detecting unit 4101, actual world estimating
unit 4102, and image generating unit 4103. While not shown in the
drawings, the signal processing device (image processing device) of
the configuration shown in FIG. 329 may be applied instead of these
data continuity detecting unit 4101, actual world estimating unit
4102, and image generating unit 4103. In this case, the comparing
unit 804 in FIG. 329 supplies the output thereof as region
identifying information to the region detecting unit 4111, and the
image generating unit 4501 supplies the output image (second
pixels) to the selector 4112.
With the above description, an example wherein the actual world is
estimated by processing image data acquired by the sensor 2
employing integration effects when processing an image, thereby
performing image processing adapted to meet the actual world has
been described.
However, light signals, which are cast upon the sensor 2, are
actually cast via an optical system made up of a lens or the like
provided immediately prior to the sensor 2. Accordingly, it is
necessary to consider influence due to the optical system when
processing the image by estimating the actual world from the image
acquired by the sensor 2.
FIG. 331 is a diagram illustrating an example of the configuration
of an optical system (optical block 5110) provided at the previous
stage of the sensor 2.
An actual world light signal is cast upon an IR cut filter 5102 via
a lens 5101 of the optical block 5110. The IR cut filter removes
light components in an infrared region, of light frequency
components, which can be received by a CCD 5104 (corresponding to
the sensor 2). According to this processing, unnecessary light,
which cannot be recognized by the human eyes, is removed. Further,
the light signal is cast upon an OLPF (Optical Low Pass Filter)
5103 following passing through the IR cut filter 5102.
The OLPF 5103 subjects a high-frequency light signal, which changes
in a range of the pixel area or less of the CCD 5104, to smoothing
to reduce the irregularities of the amount of light being cast upon
within the area of one pixel of the CCD 5104.
Accordingly, in order to consider the influence due to the optical
block 5110, it is necessary to consider the influence due to the
processing performed by the IR cut filter 5102 and the OLPF 5103
respectively. Incidentally, this IR cut filter 5102 and the OLPF
5103 make up an integral-type filter 5112 as shown in FIG. 332, and
accordingly, mounting and detaching thereof is sometimes performed
in an integral manner. Also, the influence due to the IR cut filter
5102 can be suppressed by providing a filter 5111 which passes
through short-wave light alone for example, as shown in FIG.
332.
Now, description will be made regarding image processing, which
takes the influence due to the OLPF 5103 into consideration.
The OLPF 5103 is, as shown in FIG. 333, provided with two liquid
crystals 5121a and 5121b, and a phase plate 5122 such as sandwiched
by the two liquid crystals 5121a and 5121b.
The liquid crystal plates 5121a and 5121b, as shown in FIG. 334,
each of which thickness is t, are set with a crystal axis having a
predetermined angle as to the approach direction of light. Upon
light with this angle being cast upon the liquid crystal plate
5121a in the z direction, the incident light is decomposed into a
normal ray in the same direction as the incident light and an
abnormal ray with a predetermined angle as to the incident light
respectively, and are emitted to the crystal 5121b of the
subsequent stage with a certain interval d (in the x direction). At
this time, the liquid crystal plate 5121a extracts two types of
light having a different-angle waveform, which are mutually
different 90 degrees, and emits these two types of light as a
normal ray (e.g., waveform in the y direction) L1 and an abnormal
ray L2 (e.g., waveform in the x direction).
The phase plate 5122 (not shown in FIG. 334) allows each of the
waveform of a normal ray and abnormal ray to pass through, and also
generates light having a waveform perpendicular to the waveform
thereof to emit this to the liquid crystal plate 5121b. That is to
say, in this case, the phase plate 5122 allows the waveform of the
incident normal ray to pass through and also generates a waveform
in the x direction since the incident normal ray has a waveform in
the y direction, and on the other hand, with regard to an abnormal
ray, the phase plate 5122 allows the incident abnormal ray itself
to pass through and also generates a waveform in the y direction
different from the waveform thereof 90 degrees since the incident
abnormal ray has a waveform in the x direction when being cast
thereupon, and emits both rays to the crystal plate 5121b.
The crystal plate 5121b decomposes each of the incident normal ray
L1 and abnormal ray L2 into normal rays and abnormal rays (L1 and
L3, and L2 and L4) at the incident positions, output these such
that the mutual distance becomes d. As a result, as shown in FIG.
335, for example, the light L1 cast from the backside of a paper is
decomposed into light L1 and L2 by the liquid crystal 5121a
respectively, and further, decomposed into L1 and L3, and L2 and L4
by the liquid crystal 5121b respectively. Note that at this time,
light energy is decomposed into a half at one time decomposition,
and accordingly, the OLPF 5103 outputs the incident light while
dispersing the incident light into positions apart by a distance d
(referred to as OLPF amount-of-movement d as well) with a
proportion of 25% in the horizontal direction and in the vertical
direction. As a result, light for the worth of different four
pixels, which are superimposed by 25% respectively, is received at
each pixel of the CCD 5104, and converted into pixel values,
thereby generating image data.
This OLPF amount-of-movement d is obtained with the following
Expression (248).
d=t.times.(n.sub.e.sup.2-n.sub.o.sup.2)/(2.times.n.sub.e.times.n.s-
ub.o) (248)
Note that the OLPF 5103 is not restricted to dispersing the
incident light into four pixels as described above, rather, may
disperse the incident light into the number of pixels other than
that using a greater number of crystals.
Thus, the incident light cast upon the sensor 2 is changed from
that in the actual world by the optical block 5110. Now,
description will be made regarding the processing of image data,
which takes the properties of the above optical block 5110 into
consideration (particularly takes the properties of the OLPF 5103
into consideration, here).
FIG. 336 is a block diagram illustrating the configuration of a
signal processing device, which is configured so as to process
image data taking the properties of the above optical block 5110
into consideration. Note that the components having the same
configurations as those described with reference to FIG. 3 are
appended with the same reference numerals, and the description
thereof is omitted as appropriate.
An OLPF removing unit 5131, which particularly takes the properties
of the OLPF 5103, of the above optical block 5110 included in the
input image into consideration, converts (estimates) the input
image into an image which is to be cast upon the optical block
5110, and outputs the converted image to the data continuity
detecting unit 102 and the actual world estimating unit 102.
Next, description will be made regarding the configuration of the
OLPF removing unit 5131 shown in FIG. 336 with reference to FIG.
337.
A class tap extracting unit 5141 extracts the pixel values of
multiple pixels (e.g., nine pixels in total, which are adjacent to
the horizontal direction, vertical direction, or
upper/lower/left/right oblique direction, including the pixel of
interest, such as shown in FIG. 338. Note that in FIG. 338, the
pixel of interest is represented with a double circle, and the
other pixels are represented with a circle) in positions
corresponding to the pixels of the input image data as class taps,
and outputs these to the features computing unit 5142.
The features computing unit 5142 computes features based on the
pixel values of a class tap input from the class tap extracting
unit 5141, and outputs the result to a class classification unit
5143. For example, examples of the features include the sum of the
pixel values of the pixels of a class tap, and the sum of
difference between adjacent pixels.
The class classification unit 5143 determines the class (class
code) of each pixel based on features input from the features
computing unit 5142, extracts the determined class information to a
prediction tap extracting unit 5145, and also controls coefficient
memory 5144 to supply the prediction coefficient corresponding to
the determined class to a pixel value computing unit 5146. This
class is, in the event that features are the sum of adjacent pixels
for example, set according to a range of the value to become the
sum thereof. For example, a class code is set such that class 1 in
the event that the sum thereof is 0 through 10, and class 2 in the
event that the sum thereof is 11 through 20.
The prediction coefficient for each class code based on features,
which is stored in the coefficient memory 5144, is computed by
learning processing using a later-described learning device 5150
beforehand with reference to FIG. 341, and stored.
The prediction tap extracting unit 5145 extracts the pixel values
of multiple pixels serving as a prediction tap (sometimes identical
to a class tap) corresponding to the pixel of interest in the input
image based on the class information input from the class
classification unit 5143, and outputs the extracted pixel values to
the pixel value computing unit 5146. Prediction taps are set for
each class, for example, the pixel of interest alone in the case of
class 1, 3 pixels.times.3 pixels centered on the pixel of interest
in the case of class 2, and 5 pixels.times.5 pixels centered on the
pixel of interest in the case of class 3.
The pixel value computing unit 5146 computes the pixel values based
on the pixel values of the pixels serving as a prediction tap input
from the prediction tap extracting unit 5145, and the prediction
coefficient value supplied from the coefficient memory 5144,
generates an output image based on the computed pixel values, and
outputs this. The pixel value computing unit 5146 obtains (predicts
and estimates) the pixels of a predicted image by executing a
product arithmetic operation shown in the following Expression
(249), for example.
'.times..times. ##EQU00152##
In Expression (249), q' represents the pixel of the predicted image
(an image predicted from a student image). Each of c.sub.i (i
represents an integer value of 1 through n) represents the
corresponding prediction tap. On the other hand, each of d.sub.i
represents the corresponding prediction coefficient.
As described above, the OLPF removing unit 5131 predicts and
estimates an image obtained by removing the influence due to the
OLPF as to the input image from the input image.
Next, description will be made regarding signal processing by the
signal processing device described with reference to FIG. 336, with
reference to the flowchart in FIG. 339. Note that the processing of
steps S5102 through S5104 in the flowchart shown in FIG. 339 is the
same as the processing described with reference to the flowchart in
FIG. 40, so description thereof will be omitted.
In step S5101, the OLPF removing unit 5131 executes the processing
for removing OLPF.
Now, the processing for removing OLPF will be described with
reference to the flowchart in FIG. 340.
In step S5011, the class tap extracting unit 5141 extracts a class
tap regarding each pixel of the input image, and outputs the pixel
values of the pixels of the extracted class tap to the features
computing unit 5142.
In step S5012, the features computing unit 5142 computes
predetermined features based on the pixel values of the pixels of
the class tap input from the class tap extracting unit 5141, and
outputs these to the class classification unit 5143.
In step S5013, the class classification unit 5143 classifies a
class based on the features input from the features computing unit
5142, and output the classified class code to the prediction tap
extracting unit 5145.
In step S5014, the prediction tap extracting unit 5145 extracts the
pixel values of multiple pixels serving as a prediction tap from
the input image based on the class code information input from the
class classification unit 5143, and outputs the extracted pixel
values to the pixel value computing unit 5146.
In step S5015, the class classification unit 5143 controls the
coefficient memory 5144 to read out the corresponding prediction
coefficient according to the classified class (class code) to the
pixel value computing unit 5146.
In step S5016, the pixel value computing unit 5146 computes pixel
values based on the pixel values of the pixels serving as a
prediction tap input from the prediction tap extracting unit 5145,
and the prediction coefficient supplied from the coefficient memory
5144.
In step S5017, the pixel value extracting unit 5146 determines
regarding whether or not the pixel values regarding all of the
pixels have been computed, and in the event that determination is
made that the pixel values regarding all of the pixels have not
been computed, the processing returns to step S5011. That is to
say, the processing of steps S5011 through S5017 is repeated until
determination is made that the pixel values regarding all of the
pixels have been computed.
In step S5017, in the event that determination is made that the
pixel values regarding all of the pixels have been computed, the
pixel value computing unit 5146 outputs the computed image.
According to the above arrangement, it becomes possible to remove
the influence as to the image generated by the OLPF 5103 generated
by the optical block 5110.
Next, description will be made regarding the learning device 5150
which learns prediction coefficients to be stored in the
coefficient memory 5144 shown in FIG. 337 beforehand with reference
to FIG. 341.
The learning device 5150 generates a student image and a tutor
image, which are made up of an image with the standard resolution,
using a high-resolution image serving as an input image, and
executes learning processing. Note that images with the standard
resolution will be referred to as "SD (Standard Definition) image"
hereafter as appropriate. Also, pixels forming the SD image will be
referred to as "SD pixels" as appropriate. Alternately, on the
other hand, high-resolution images will be referred to as "HD (High
Definition) image" hereafter as appropriate. Also, pixels forming
the HD image will be referred to as "HD pixels" as appropriate.
Further, a class tap extracting unit 5162, features computing unit
5163, and prediction tap extracting unit 5165 of a learning unit
5152 are the same as the class tap extracting unit 5141, features
computing unit 5142, and prediction tap extracting unit 5145 of the
OLPF removing unit 5131 shown in FIG. 337, so description thereof
will be omitted.
A student image generating unit 5151 converts an HD image serving
as an input image into an SD image taking the OLPF 5103 into
consideration, generates a student image optically influenced by
the OLPF 5103, and outputs this to image memory 5161 of the
learning unit 5152.
The image memory 5161 at the learning unit 5152 temporarily stores
the student image made up of the SD image, and then outputs this to
the class tap extracting unit 5162, and the prediction tap
extracting unit 5165.
The class classification unit 5164 outputs the classified result
(class code described above) of a class for each pixel input from
the features extracting unit 5163 to the prediction tap extracting
unit 5165, and learning memory 5167.
A supplementing computing unit 5166 generates the summation term of
each term necessary for generating a later-described normal
equation from the pixel values of the pixels of a prediction tap
input from the prediction tap extracting unit 5165 and the pixel
values of the pixels of an image input from a tutor image
generating unit 5153 with supplement, and outputs this to the
learning memory 5167.
The learning memory 5167 stores a class code supplied from the
class classification unit 5164 and the supplemented result input
from the supplementing computing unit 5166, which are correlated
with each other, and supplies these to a normal equation computing
unit 5168 as appropriate.
The normal equation computing unit 5168 generates a normal equation
based on the class codes stored in the learning memory 5167 and the
supplemented result, and also computes the normal equation to
obtain each prediction coefficient, and then stores each obtained
prediction coefficient, which is correlated with the corresponding
class code in the coefficient memory 5154. Note that the prediction
coefficient stored in this coefficient memory 5154 is to be stored
in the coefficient memory of the OLPF removing unit 5131 shown in
FIG. 337.
Description will be made more in detail regarding the normal
equation computing unit 5168.
In the above Expression (249), each of the prediction coefficients
d.sub.i is undetermined before learning. The learning processing is
performed by inputting the multiple pixels of the tutor image for
each class code. If we say that there are m pixels of the tutor
image corresponding to a certain class code, and each of the m
pixels of the tutor image is described as q.sub.k (k represents an
integer value of 1 through m), the following Expression (250) is
introduced from the Expression (249).
.times..times. ##EQU00153##
That is to say, the Expression (250) indicates that the pixel
q.sub.k of a certain tutor image can be predicted and estimated by
computing the right side thereof. Note that in Expression (250),
e.sub.k represents an error. That is to say, the pixel q.sub.k' of
the predicted image (image obtained by performing prediction
computation from a student image) serving as computation results
obtained by computing the right side does not exactly match the
actual pixel q.sub.k of the tutor image, but contains a certain
error e.sub.k.
Accordingly, in Expression (250), the prediction coefficients
d.sub.i which exhibit the minimum of the sum of the squares of
errors e.sub.k should be obtained by the learning processing, for
example.
Specifically, the number of the pixels q.sub.k of the tutor image
prepared for the learning processing should be greater than n
(i.e., m>n). In this case, the prediction coefficients d.sub.i
are determined as a unique solution using the least square
method.
That is to say, the normal equations for obtaining the prediction
coefficients d.sub.i in the right side of the Expression (250)
using the least square method are represented by the following
Expression (251).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..function.
.times..times..times..times..times..times..times..times.
##EQU00154##
Accordingly, the normal equations represented by the Expression
(251) are created and solved, thereby determining the prediction
coefficients d.sub.i as a unique solution.
Specifically, let us say that the matrices in the Expression (251)
representing the normal equations are defined as the following
Expressions (252) through (254). In this case, the normal equations
are represented by the following Expression (255).
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times.
##EQU00155##
As shown in Expression (253), each component of the matrix
D.sub.MAT is the prediction coefficient d.sub.i which is to be
obtained. Accordingly, in Expression (255), in the event that the
matrix C.sub.MAT in the left side and the matrix Q.sub.MAT in the
right side are determined, the matrix D.sub.MAT (i.e., the
prediction coefficients d.sub.i) can be obtained using matrix
computation.
More specifically, as shown in Expression (252), each component of
the matrix C.sub.MAT can be computed as long as the prediction tap
c.sub.ik is known. With the present embodiment, the prediction tap
c.sub.ik is extracted by the prediction tap extracting unit 5165,
whereby the supplementing computing unit 5166 can supplement each
component of the matrix C.sub.MAT using the prediction tap c.sub.ik
supplied from the prediction tap extracting unit 5165.
Also, with the present embodiment, each component of the matrix
Q.sub.MAT can be computed as shown in Expression (254) as long as
the prediction tap c.sub.ik and the pixel q.sub.k of the tutor
image are known. Note that the prediction tap C.sub.ik is the same
as that included in each component of the matrix C.sub.MAT, and the
pixel q.sub.k of the tutor image is the SD pixel of the tutor image
corresponding to the pixel of interest (SD pixel of the student
image). Accordingly, the supplementing computing unit 5166 can
supplement each component of the matrix Q.sub.MAT based upon the
prediction tap c.sub.ik supplied from the prediction tap extracting
unit 5165 and the tutor image.
Thus, the supplementing computing unit 5166 computes each component
of the matrix C.sub.MAT and the matrix Q.sub.MAT, correlates the
computed result with the corresponding class code, and stores this
in the learning memory 5167.
The normal equation computing unit 5168 generates a normal equation
corresponding to the class codes stored in the learning memory
5167, and computes the prediction coefficient d.sub.i serving as
each component of the matrix D.sub.MAT in the above Expression
(255).
Specifically, the above Expression (255) can be transformed into
the following Expression (256).
.times. ##EQU00156##
In Expression (256), each component of the matrix D.sub.MAT in the
left side is the prediction coefficient d.sub.i which is to be
obtained. On the other hand, each component of the matrix C.sub.MAT
and the matrix Q.sub.MAT is supplied from the learning memory 5167.
With the present embodiment, upon reception of each component of
the matrix C.sub.MAT and the matrix Q.sub.MAT corresponding to a
certain class code stored in the learning memory 5167, the normal
equation computing unit 5168 executes the matrix computation
represented by the right side of Expression (255), thereby
computing the matrix D.sub.MAT. Then, the normal equation computing
unit 5168 stores the computation results (prediction coefficient
d.sub.i) in the coefficient memory 5154 in association with the
class code.
Next, description will be made regarding a student image and tutor
image employed for learning based on the relationship between the
OLPF removing unit 5131 and learning unit 5131 in FIG. 337
described above.
As shown in FIG. 342, the learning unit 5152 obtains a prediction
coefficient by learning using an image subjected to filter
processing by the OLPF 5103 (hereafter, referred to as image with
OLPF) and an image not subjected to filter processing (hereafter,
referred to as image without OLPF).
The OLPF removing unit 5131 converts an image with OLPF into an
image from which the influence of the filtering processing by the
OLPF 5103 is removed (hereafter, referred to as OLPF-removed image)
using the prediction coefficient obtained by learning with the
learning unit 5152 (processing described with reference to the
flowchart shown in FIG. 339).
That is to say, as shown in FIG. 343, the learning processing
performed at the learning unit 5152 is executed using a learning
pair made up of a tutor image serving as an image with OLPF, and a
student image serving as an image without OLPF.
Accordingly, a learning pair is made up by generating an image in
the case of receiving the incident light at the sensor 2 in a state
wherein the OLPF is provided, and an image in the case of receiving
the incident light at the sensor 2 in a state wherein the OLPF is
nor provided, but it is actually extremely difficult to use each
image by accurately positioning each image in increments of
pixels.
In order to solve this problem, the learning device 5110 generates
an image with OLPF and an image without OLPF using a
high-resolution image serving as an input image by means of
simulation.
Now, description will be made regarding a method for generating a
tutor image using a tutor image generating unit 5153 in the
learning device 5110, and a method for generating a student image
using a student image generating unit 5151.
FIG. 344 is a block diagram illustrating the detailed configuration
of the tutor image generating unit 5153 and the student image
generating unit 5151 of the learning device 5110.
A 1/16 average processing unit 5153a of the tutor image generating
unit 5153 obtains the average pixel value of pixel values of 16
pixels in total of 4 pixels.times.4 pixels in the entire range of a
high-resolution image serving as an input image, replaces all of
the pixel values of the 16 pixels with the obtained average pixel
value to generate and output a tutor image. According to this
processing, the number of pixels of the HD image becomes 1/16
pixels (1/4 pixels each in the horizontal direction and in the
vertical direction) in appearance.
That is to say, this 1/16 average processing unit 5153a regards
each pixel of the HD image serving as an input image as the light
cast upon the sensor 2, and regards the range of 4 pixels.times.4
pixels of the HD image as one pixel of the SD image, thereby
generating a kind of spatial integration effects, and virtually
generating an image (image without OLPF), which is to be generated
at the sensor 2, with no influence due to the OLPF 5103.
An OLPF simulation processing unit 5151a of the student image
generating unit 5151 disperses the pixel values of the pixels of
the HD image which is input in increments of 25%, and superimposes
these, as described with reference to FIG. 334 and FIG. 335,
thereby simulating operation caused due to the OLPF 5103 when
viewing each pixel of the HD image as light.
A 1/16 average processing unit 5135b is the same as the 1/16
average processing unit 5153a of the tutor image generating unit
5153, replaces all of the pixel values of the 16 pixels with the
average pixel value of the 16 pixels in total of 4 pixels.times.4
pixels, and generates a student image made up of an SD image.
More particularly, all of the pixels are subjected to the
processing wherein the OLPF simulation processing unit 5151a
disperses a value obtained by dividing the pixel value of a pixel
P1 at the incident position into pixels P1 through P4 respectively
for example as shown in FIG. 345, and then pixel values are
obtained by superimposing the values dispersed respectively.
According to this processing, for example, the pixel P4 shown in
FIG. 345 becomes the average pixel value of the pixels P1 through
P4.
In FIG. 345, each grid corresponds to one pixel of an HD image.
Also, 4 pixels.times.4 pixels surrounded by a dotted line
correspond to one pixel of an SD image.
That is to say, in FIG. 345, the distance between the pixels P1 and
P2, the distance between the pixels P1 and P3, and the distance
between the pixels P2 and P4 are equivalent to an
amount-of-movement d by the OLPF 5103 shown in FIG. 335.
The reason why the distance between the pixels P1 and P2, the
distance between the pixels P1 and P3, and the distance between the
pixels P2 and P4, become 2 pixels, is that the OLPF
amount-of-movement d by the OLPF 5103 is actually 3.35 .mu.m, but
on the other hand, the pixel pitch of the CCD 5104 (the widths
between pixels in the horizontal direction and in the vertical
direction) is actually 6.45 .mu.m, and the relative ratio thereof
is 1.93, as shown in FIG. 346. That is to say, the OLPF
amount-of-movement is set to 2 pixels such as surrounded with the
dotted line in the drawing to set the pixel pitch to 4 pixels, and
consequently, the relative ratio thereof becomes 2.0, and
accordingly, an event occurred by the OLPF 5103, which is to be
cast upon the sensor 2, can be simulated in a state similar to an
actual measured value of 1.93.
Similarly, as shown in FIG. 346, an arrangement may be made wherein
the OLPF amount-of-movement is set to 4 pixels, and the pixel pitch
is set to 8 pixels, i.e., the other OLPF amount-of-movement and
pixel pitch may be employed as long as the OLPF amount-of-movement
and pixel pitch are set while keeping this proportion. Further,
even if the OLPF amount-of-movement is set to 6 pixels, and the
pixel pitch is set to 11 pixels, the relative ratio thereof can
keep 1.83, the processing simulation with this proportion may be
performed.
In the event that the tutor image generating unit 5153 generates an
image such as shown in FIG. 347, the student image generating unit
5151 generates an image such as shown in FIG. 348. Both images are
displayed in a mosaic pattern since 4 pixels.times.4 pixels of the
HD image are essentially displayed as a single pixel of the SD
image, but with the tutor image shown in FIG. 347, the edge portion
shown in a white color is displayed more clearly than that in the
student image shown in FIG. 348, and accordingly, an image caused
by the influence of the OLPF 5103 is generated upon the student
image.
Next, description will be made regarding the learning processing
with reference to the flowchart shown in FIG. 349.
In step S5031, the OLPF simulation processing unit 5151a of the
student image generating unit 5151, as described with reference to
FIG. 345, disperses the pixel values of the pixels of the HD image
which is input into four pixels in increments of 25%, generates
pixel values by superimposing the pixel values dispersed at each
pixel position, simulates the operation caused by the OLPF 5103,
and outputs the processed results to the 1/16 average processing
unit 5151b.
In step S5032, the 1/16 average processing unit 5151b obtains an
average pixel value in increments of 16 pixels in total of 4
pixels.times.4 pixels regarding the image subjected to the OLPF
simulation processing input from the OLPF simulation processing
unit 5151a, further replaces the pixel values of the 16 pixels with
the average value thereof in order, generates a student image,
which becomes an SD image in appearance, and outputs this to the
image memory 5161 of the learning unit 5152.
In step S5033, the class tap extracting unit 5162 extracts the
pixel value of a pixel serving as the class tap of a pixel of
interest from the image data stored in the image memory 5161, and
outputs the extracted pixel value of the pixel to the features
computing unit 5163.
In step S5034, the features extracting unit 5163 computes the
features corresponding to the pixel of interest using the pixel
value information of the pixel of the class tap input from the
class tap extracting unit 5162, and outputs the computed features
to the class classification unit 5164.
In step S5035, the class classification unit 5164 classifies the
class corresponding to the pixel to become a pixel of interest to
determine a class code based on the features input, outputs this to
the prediction tap extracting unit 5165, and also stores this to
the learning memory.
In step S5036, the prediction tap extracting unit 5165 extracts the
pixel value information of the pixel of the prediction tap
corresponding to the pixel of interest of the image data stored in
the image memory 5161 based on the class code input from the class
classification unit 5164, and outputs this to the supplementing
computing unit 5166.
In step S5037, the 1/16 average processing unit 5153a of the tutor
image generating unit 5153 obtains an average pixel value in
increments of 16 pixels in total of 4 pixels.times.4 pixels
regarding an HD image serving as an input image, replaces the pixel
values of the 16 pixels with the obtained average pixel value,
thereby generating an image without OLPF (SD image in appearance),
which is not influenced due to the OLPF 5103, to output this to the
supplementing computing unit 5166.
In step S5038, the supplementing computing unit 5166 supplements a
value to become the summation of each term of a normal equation
based on the pixel values of the pixels of the tutor image input
from the tutor image generating unit 5153, outputs the supplemented
result to the learning memory 5167, and stores this in association
with the corresponding class code.
In step S5039, the normal equation computing unit 5168 determines
regarding whether or not the supplementing processing regarding all
the pixels of the input image has been completed, and in the event
that determination is made that the supplementing processing
regarding all the pixels of the input image has not been completed,
the processing returns to step S5032, wherein the subsequent
processing is repeated. In other words, the processing of steps
S5032 through S5039 is repeated until the supplementing processing
regarding all the pixels of the input image has been completed.
In the event that determination is made that the supplementing
processing regarding all the pixels of the input image has been
completed in step S5039, the normal equation computing unit 5168
computes a normal equation while correlated with the corresponding
class code based on the supplemented results stored in the learning
memory 5167, obtains a prediction coefficient thereof to output
this to the coefficient memory 5154.
In step S5041, the normal equation computing unit 5168 determines
regarding whether or not the computation for obtaining prediction
coefficients as to all of the classes has been completed, and in
the event that determination is made that the computation for
obtaining prediction coefficients as to all of the classes has not
been completed, the processing returns to step S5040. In other
words, the processing of step S5040 is repeated until the
computation for obtaining prediction coefficients as to all of the
classes has been completed.
In step S5041, in the event that determination is made that the
computation for obtaining prediction coefficients as to all of the
classes has been completed, the processing thereof ends.
According to the above learning processing, the OLPF removing unit
5131 can generate an image similar to an actual world image of
which the OLPF processing effects are removed from the input image
subjected to the filtering processing by the OLPF 5103 by using the
prediction coefficients stored in the coefficient memory 5154, such
as copying the prediction coefficients into the coefficient memory
5144, or the like.
For example, by employing thus obtained prediction coefficients, in
the event that an image subjected to the filtering processing by
the OLPF 5103 (image obtained by simulating the processing by the
OLPF 5103) such as shown in FIG. 348 is input, the OLPF removing
unit 5131 generates an image such as shown in FIG. 350 using the
OLPF removing processing described with reference to the flowchart
shown in FIG. 340.
It can be understood that the image thus processed, which is shown
in FIG. 350, is generally the same image as the input image not
subjected to the filtering processing by the OLPF 5103 such as
shown in FIG. 347.
Also, as shown in FIG. 351, it can be understood that an image of
which the effects by the OLPF are removed exhibits a value closer
to an image not subjected to the effects by the OLPF than an image
subjected to the filtering processing by the OLPF even in
comparison of change in the pixels in the x direction at the
certain same position in the y direction of the images in FIG. 347,
FIG. 348, and FIG. 350.
Note that in FIG. 351, a solid line represents change in the pixel
values corresponding to the image (image without OLPF) shown in
FIG. 347, a dotted line represents the image (image with OLPF)
shown in FIG. 348, and a single-dot broken line represents the
image (OLPF removed image) shown in FIG. 350.
According to the above arrangement, image data wherein the real
world light signals are cast upon multiple pixels each having
spatial integration effects via the optical low pass filter is
acquired, the light signals cast upon the optical low pass filter
are estimated so as to take that the light signals are dispersed
and integrated in at least one-dimensional direction of the spatial
directions by the optical low pass filter into consideration, and
accordingly, it becomes possible to obtain more accurate and
higher-precision processed results as to events in the real world
taking the real world wherein the data is acquired into
consideration.
With the above examples, description has been made regarding
examples wherein the influence of the filtering processing by the
OLPF 5103 is removed at the previous stage of the data continuity
detecting unit 101, but the actual world may be estimated using the
actual world estimating unit 102 taking the influence by the OLPF
5103 into consideration. Accordingly, in this case, the
configuration of the signal processing device becomes the
configuration described with reference to FIG. 3.
FIG. 352 is a block diagram illustrating the configuration of the
actual world estimating unit 102 so as to estimate the actual world
taking the influence by the OLPF 5103 into consideration.
As shown in FIG. 352, the actual world estimating unit 102 includes
a condition setting unit 5201, input image storing unit 5202, input
pixel value acquiring unit 5203, integration component computing
unit 5204, normal equation generating unit 5205, and approximation
function generating unit 5206.
The condition setting unit 5201 sets a pixel range (tap range) used
for estimating the function F(x, y) corresponding to a pixel of
interest, and the number of dimensions n of the approximation
function f(x, y), g(x, y).
The input image storing unit 5202 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel value acquiring unit 5203 acquires, of the input
images stored in the input image storing unit 5202, an input image
region corresponding to the tap range set by the condition setting
unit 5201, and supplies this to the normal equation generating unit
5205 as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described. Note that
a specific example of the input pixel value table will be described
later.
Incidentally, as described with reference to FIG. 344 and FIG. 345,
the OLPF 5103 disperses the incident light into four points with
the OLPF amount-of-movement d. Accordingly, with the pixels on the
image, the pixel values thereof are generated by each 25% of the
pixel values at the four points including the own pixel position
being superimposed. Note that FIG. 353 illustrates that the ranges
surrounded with a dotted line represent different four pixel
points, and 25% of each is superimposed.
As described above, the incident light is dispersed into four
points such as shown in FIG. 354 by the OLPF 5103, an approximation
function g (x, y) indicating the dispersed light distribution
immediately prior to the sensor 2 becomes a relational expression
such as shown in the following Expression (257) using the
approximation function f (x, y) which approximates the actual
world. Note that FIG. 354 illustrates curves having a convex shape
on the top thereof represent the approximation function f (x, y),
and the approximation function wherein these curves are dispersed
into four curves, and then superimposed is g (x, y).
g(x,y)=f(x,y)+f(x-d,y)+f(x,y-d)+f(x-d,y-d) (257)
Also, the approximation function f (x, y) of the actual world is
represented with the following Expression (258).
.function..times..times..times. ##EQU00157##
Here, w.sub.i represents the coefficients of the approximation
function, and s (=cot .theta.: .theta. is continuity angle)
represents a gradient as continuity.
Accordingly, the approximation function g (x, y) indicating the
light distribution immediately prior to the sensor 2 is represented
with the following Expression (259).
.function..times..times..times..times..times..times..times..times..times.-
.times..times..times. ##EQU00158##
The actual world estimating unit 102 computes the features w.sub.i
of the approximation function f (x, y), as described above.
Expression (259) can be expressed as in the following Expression
(260).
.times..intg..times..intg..times..function..times.d.times.d.times..intg..-
times..intg..times..times..times..times..times..times..times..times..times-
..times..times..times..times..times..times.d.times.d.times..intg..times..i-
ntg..times..times..times..times..times.d.times.d.times..intg..times..intg.-
.times..times..times..times..times.d.times.d.times..intg..times..intg..tim-
es..times..times..times..times.d.times.d.times..intg..times..intg..times..-
times..times..times..times.d.times.d.times..times..function..times..times.-
.times..times..times..times..times..times..times..times..function..times..-
times..times..times..times..times..times..times..times..times..function..t-
imes..times..times..times..times..times..times..times..times..times..funct-
ion..times..times..times..times..times..times..times..times..times..times.-
.times..function..times..times..times..times..times..times..times..times..-
times..times..times..times..times..times..times..times..times..times..time-
s..times..times..times..times..times..times..times..times..times..times..t-
imes..times..function. ##EQU00159##
In Expression (260), S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5)
represents the integral components of i-dimensional terms. That is
to say, the integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) are as shown in the following Expression (261).
.function..times..times..times..times..times..times..times..times..functi-
on..times. ##EQU00160##
The integration component calculation unit 5204 computes the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5).
Specifically, the integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) shown in Expression (261) can be computed as long as the
relative pixel positions (x, y), the gradient s and i of
i-dimensional terms are known. Of these, the relative pixel
positions (x, y) are determined with a pixel of interest, and a tap
range, the variable s is cot .theta., which is determined with the
angle .theta., and the range of i is determined with the number of
dimensions n respectively.
Accordingly, the integration component computing unit 5204 computes
the integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) based
on the tap range and the number of dimensions set by the condition
setting unit 5201, and the angle .theta. of the data continuity
information output from the data continuity detecting unit 101, and
supplies the calculated results to the normal equation generating
unit 5205 as an integration component table.
The normal equation generating unit 5205 generates a normal
equation in the case of obtaining the above Expression (260) by the
least square method using the input pixel value table supplied from
the input pixel value acquiring unit 5203, and the integration
component table supplied from the integration component computing
unit 5206, and outputs this to the approximation function
generating unit 5206 as a normal equation table. Note that a
specific example of a normal equation will be described later.
The approximation function generating unit 5206 computes the
respective features w.sub.i of the above Expression (259) (i.e.,
the coefficients w.sub.i of the approximation function f(x, y)
serving as a two-dimensional polynomial) by solving the normal
equation included in the normal equation table supplied from the
normal equation generating unit 5205 using the matrix solution, and
output these to the image generating unit 103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40), which
takes the influence by the OLPF 5103 into consideration, with
reference to the flowchart in FIG. 355.
For example, let us say that the light signal in the actual world 1
having continuity in the spatial direction represented with the
gradient G.sub.F has been detected by the sensor 2, and has been
stored in the input image storing unit 5202 as an input image
corresponding to one frame. Also, let us say that the data
continuity detecting unit 101 has output the angle .theta. as data
continuity information, of the input image.
In this case, in step S5201, the condition setting unit 5201 sets
conditions (a tap range and the number of dimensions).
For example, let us say that a tap range 5241 shown in FIG. 356 has
been set, and also 5 has been set as the number of dimensions.
FIG. 356 is a diagram for describing an example of a tap range. In
FIG. 356, the X direction and Y direction represent the X direction
and Y direction of the sensor 2. Also, the tap range 5241
represents a pixel group made up of 20 pixels (20 squares in the
drawing) in total of 4 pixels in the X direction and also 5 pixels
in the Y direction.
Further, as shown in FIG. 356, let us say that a pixel of interest
has been set to a pixel, which is the second pixel from the left
and also the third pixel from the bottom in the drawing, of the tap
range 5241. Also, let us say that each pixel is denoted with a
number l such as shown in FIG. 356 (l is any integer value of 0
through 19) according to the relative pixel positions (x, y) from
the pixel of interest (a coordinate value of a pixel-of-interest
coordinates system wherein the center (0, 0) of the pixel of
interest is taken as the origin).
Now, description will return to FIG. 355, wherein in step S5202,
the condition setting unit 5201 sets a pixel of interest.
In step S5203, the input pixel value acquiring unit 5203 acquires
an input pixel value based on the condition (tap range) set by the
condition setting unit 5201, and generates an input pixel value
table. That is to say, in this case, the input pixel value
acquiring unit 5203 generates a table made up of 20 input pixel
values P (l) as an input pixel value table.
Note that in this case, the relation between the input pixel values
P (l) and the above input pixel values P (x, y) is a relation shown
in the following Expression (262). However, in Expression (262),
the left side represents the input pixel values P (l), and the
right side represents the input pixel values P (x, y).
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function..times..times..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..times..t-
imes..function..function..times..times..function..function..times..times..-
function..function..times..times..function..function..times..times..functi-
on..function..times..times..function..function..times..times..function..fu-
nction..times..times..function..function..times..times..function..function-
. ##EQU00161##
In step S5204, the integration component computing unit 5204
computes integral components based on the conditions (a tap range
and the number of dimensions) set by the condition setting unit
5201, and the data continuity information (angle .theta.) supplied
from the data continuity detecting unit 101, and generates an
integration component table.
In this case, as described above, the input pixel values are not P
(x, y) but P (l), and are acquired as the value of a pixel number
l, so the integration component computing unit 5204 computes the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) in the
above Expression (261) as a function of l such as the integral
components S.sub.i (l) shown in the left side of the following
Expression (263). S.sub.i(l)=S.sub.i(x-0.5,x+0.5,y-0.5,y+0.5)
(263)
Specifically, in this case, the integral components S.sub.i (l)
shown in the following Expression (264) are computed.
.function..function..times..times..function..function..times..times..func-
tion..function..times..times..function..function..times..times..function..-
function..times..times..function..function..times..times..function..functi-
on..times..times..function..function..times..times..function..function..ti-
mes..times..function..function..times..times..function..function..times..t-
imes..function..function..times..times..function..function..times..times..-
function..function..times..times..function..function..times..times..functi-
on..function..times..times..function..function..times..times..function..fu-
nction..times..times..function..function..times..times..function..function-
. ##EQU00162##
Note that in Expression (264), the left side represents the
integral components S.sub.i (l), and the right side represents the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5). That is
to say, in this case, i is 0 through 5, and accordingly, the 120
S.sub.i (l) in total of the 20 S.sub.0 (l), 20 S.sub.1 (l), 20
S.sub.2 (l), 20 S.sub.3 (l), 20 S.sub.4 (l), and 20 S.sub.5 (l) are
computed.
More specifically, first the integration component computing unit
5204 calculates cot .theta. corresponding to the angle .theta.
supplied from the data continuity detecting unit 101, and takes the
computed result as a variable s. Next, the integration component
computing unit 5204 computes each of the 20 integral components
S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) shown in the right side of
Expression (264) regarding each of i=0 through 5 using the computed
variable s. That is to say, the 120 integral components S.sub.i
(x-0.5, x+0.5, y-0.5, y+0.5) are computed. Note that with this
calculation of the integral components S.sub.i (x-0.5, x+0.5,
y-0.5, y+0.5), the above Expression (261) is used. Subsequently,
the integration component computing unit 5204 converts each of the
computed 120 integral components S.sub.i (x-0.5, x+0.5, y-0.5,
y+0.5) into the corresponding integral components S.sub.i (l) in
accordance with Expression (264), and generates an integration
component table including the converted 120 integral components
S.sub.i (l).
Note that the sequence of the processing in step S5203 and the
processing in step S5204 is not restricted to the example in FIG.
355, the processing in step S5204 may be executed first, or the
processing in step S5203 and the processing in step S5204 may be
executed simultaneously.
Next, in step S5205, the normal equation generating unit 5205
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 5203 at the
processing in step S5203, and the integration component table
generated by the integration component computing unit 5204 at the
processing in step S5204.
Specifically, in this case, the features w.sub.i are calculated
with the least square method using the above Expression (260)
(however, in Expression (258), the S.sub.i (l) into which the
integral components S.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) are
converted using Expression (262) is used), so a normal equation
corresponding to this is represented as the following Expression
(265).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..times..function..times..function..times..function..time-
s..function..times..function..times..function. ##EQU00163##
Note that in Expression (265), L represents the maximum value of
the pixel number l in the tap range. n represents the number of
dimensions of the approximation function f(x) serving as a
polynomial. Specifically, in this case, n=5, and L=19.
If we define each matrix of the normal equation shown in Expression
(265) as the following Expressions (266) through (268), the normal
equation is represented as in the following Expression (269).
.times..function..times..function..times..function..times..function..time-
s..function..times..function..times..function..times..function..times..fun-
ction..times..function..times..function..times..function..times..function.-
.times..function..times..function..times..function..times..function..times-
..function..times..function..times..function..times..function..times..func-
tion..times..function..times..function..times. ##EQU00164##
As shown in Expression (267), the respective components of the
matrix W.sub.MAT are the features w.sub.i to be obtained.
Accordingly, in Expression (269), if the matrix S.sub.MAT of the
left side and the matrix P.sub.MAT of the right side are
determined, the matrix W.sub.MAT may be calculated with the matrix
solution.
Specifically, as shown in Expression (266), the respective
components of the matrix S.sub.MAT may be calculated with the above
integral components S.sub.i (l). That is to say, the integral
components S.sub.i (l) are included in the integration component
table supplied from the integration component computing unit 5204,
so the normal equation generating unit 5205 can calculate each
component of the matrix S.sub.MAT using the integration component
table.
Also, as shown in Expression (268), the respective components of
the matrix P.sub.MAT may be calculated with the integral components
S.sub.i (l) and the input pixel values P (l). That is to say, the
integral components S.sub.i (l) is the same as those included in
the respective components of the matrix S.sub.MAT, also the input
pixel values P (l) are included in the input pixel value table
supplied from the input pixel value acquiring unit 5203, so the
normal equation generating unit 5205 can calculate each component
of the matrix P.sub.MAT using the integration component table and
input pixel value table.
Thus, the normal equation generating unit 5205 calculates each
component of the matrix S.sub.MAT and matrix P.sub.MAT, and outputs
the calculated results (each component of the matrix S.sub.MAT and
matrix P.sub.MAT) to the approximation function generating unit
5206 as a normal equation table.
Upon the normal equation table being output from the normal
equation generating unit 5205, in step S5206, the approximation
function generating unit 5206 calculates the features w.sub.i
(i.e., the coefficients w.sub.i of the approximation function f(x,
y) serving as a two-dimensional polynomial) serving as the
respective components of the matrix W.sub.MAT in the above
Expression (269) based on the normal equation table.
Specifically, the normal equation in the above Expression (269) can
be transformed as the following Expression (270).
.times. ##EQU00165##
In Expression (270), the respective components of the matrix
W.sub.MAT in the left side are the features w.sub.i to be obtained.
The respective components regarding the matrix S.sub.MAT and matrix
P.sub.MAT are included in the normal equation table supplied from
the normal equation generating unit 5205. Accordingly, the
approximation function generating unit 5206 calculates the matrix
W.sub.MAT by calculating the matrix in the right side of Expression
(270) using the normal equation table, and outputs the calculated
results (features w.sub.i) to the image generating unit 103.
In step S5207, the approximation function generating unit 5206
determines regarding whether or not the processing of all the
pixels has been completed.
In step S5207, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S5202, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S5202 through S5207 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S5207, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
A in FIG. 357 illustrates a high-precision input image (image of a
bicycle spoke), B in FIG. 357 is an image obtained by the image of
A in FIG. 357 being subjected to the processing by the OLPF 5103, C
in FIG. 357 is an image of which pixels are generated using an
approximation function of the actual world estimated from the image
of B in FIG. 357 using the processing described with reference to
the flowchart shown in the above FIG. 355, and D in FIG. 357 is an
image generated by the image of B in FIG. 357 being generated by
the conventional class classification adaptation processing.
It can be understood that the image of C in FIG. 357 displays the
edge strongly, so that the outline of the spoke is clearly
displayed, as compared to the image of D in FIG. 357.
Also, FIG. 358 is a diagram illustrating change in the pixel value
in the horizontal direction at a certain position in the vertical
direction of the images of A through D in FIG. 357. In FIG. 358, a
single-dot broken line corresponds to the image of A in FIG. 357, a
solid line corresponds to the image of B in FIG. 357, a dotted line
corresponds to the image of C in FIG. 357, and a double-dot broken
line corresponds to the image of D in FIG. 357. As shown in FIG.
358, with around the spatial direction X=10 wherein the image of
the spoke is displayed, it can be understood that the dotted line
serving as an image which is processed by the actual world
estimating unit 102 shown in FIG. 352 taking the influence by the
OLPF 5103 into consideration can obtain a value similar to the
input image, as compared to the image, which is shown with a
double-dot broken line, generated by the conventional class
classification adaptation processing.
Particularly, a portion of which the pixel value is small is a
reflected portion of the edge portion of the spoke, but with regard
to this portion, expressiveness is improved by the processing in
which the OLPF is taken into consideration.
According to the actual world estimating unit 102 shown in FIG.
352, it becomes possible to obtain the approximation function f(x)
in the actual world in which the influence by the OLPF 5103 is
taken into consideration, and further, it becomes possible to
generate pixels in which the influence by the OLPF 5103 is taken
into consideration from the approximation function f(x) in the
actual world in which the influence by the OLPF 5103 is taken into
consideration.
As described above, as the description of the two-dimensional
polynomial approximating technique, an example wherein the
coefficients (features) w.sub.i of the approximation function f(x,
y) as to the spatial directions (X direction and Y direction) are
computed has been employed, but it is needless to say that the
one-dimensional polynomial approximating technique wherein any
one-dimensional direction alone of the spatial directions (X
direction or Y direction) is applied, may be employed as well.
According to the above arrangement, a function corresponding to the
real world light signals is estimated by estimating multiple actual
world functions assuming that the pixel value of the pixel of
interest corresponding to a position in at least one-dimensional
direction of the spatial directions of the image data acquired by
the real world light signals being cast upon multiple pixels each
having spatio-temporal integration effects via the optical low pass
filter, of which part of continuity of the real world light signals
is dropped, is the pixel value acquired by the integration in at
least one dimensional direction of the multiple actual world
functions corresponding to the optical low pass filter, thereby
enabling the actual world to be estimated in a truer manner.
With the above arrangement, the signal processing device shown in
FIG. 336 has executed the signal processing so as to remove the
influence by the OLPF 5103 from the image input from the sensor 2,
the actual world estimating unit 102 shown in FIG. 352 has
generated an actual world approximation function taking the
influence by the OLPF 5103 into consideration, and consequently,
the processing taking the influence by the OLPF 5103 into
consideration has been performed with the signal processing, but
for example, an arrangement may be made wherein an HD image without
OLPF is taken as a tutor image, and an SD image with OLPF is taken
as a student image, a prediction coefficient is set by learning,
and an image is generated with the class classification adaptation
processing.
FIG. 359 is a block diagram illustrating the configuration of a
signal processing device 5221 configured such that an HD image
without OLPF is taken as a tutor image, and an SD image with OLPF
is taken as a student image, a prediction coefficient is set by
learning, and an image is generated with the class classification
adaptation processing.
Note that the signal processing device 5221 shown in FIG. 359 is
essentially the same configuration as the OLPF removing unit 5131
shown in FIG. 337, a class tap extracting unit 5241, features
computing unit 5242, class classification unit 5243, coefficient
memory 5244, prediction tap extracting unit 5245, and pixel value
computing unit 5246 of the signal processing unit 5221 are the same
as the class tap extracting unit 5141, features computing unit
5142, class classification unit 5143, coefficient memory 5144,
prediction tap extracting unit 5145, and pixel value computing unit
5146 of the signal processing unit 5141 of the OLPF removing unit
5131, so description thereof will be omitted. However, the
prediction coefficients stored in the coefficient memory 5244 are
obtained by learning, which are different from those in the
coefficient memory 5144. Description will be made later regarding
the learning of the prediction coefficients stored in the
coefficient memory 5244 with reference to the learning device shown
in FIG. 361.
Next, description will be made regarding the signal processing by
the signal processing device 5221 shown in FIG. 359 with reference
to the flowchart shown in FIG. 360, but this processing is
essentially the same as that in the flowchart shown in FIG. 340, so
description thereof will be omitted.
According to the above arrangement, first image data acquired by
the real world light signals being cast upon multiple pixels each
having spatio-temporal integration effects via the optical low pass
filter is acquired, the multiple pixels corresponding to the pixel
of interest within second image data are extracted from the first
image data, learning is made beforehand so as to predict second
image data acquired by the light signals, which are to be cast upon
the optical low pass filter, being cast directly, based on the
first image data, and the pixel value of the pixel of interest
within the second image data is predicted based on the extracted
multiple pixels and the prediction, thereby enabling an image which
is faithful as to the actual world to be generated.
Next, description will be made regarding a learning device, which
learns (the signal processing device shown in FIG. 359 described
above serves as predicting means for predicting a pixel value using
a prediction coefficient, and accordingly, to learn prediction
coefficients means to learn the prediction means) prediction
coefficients to be stored in the coefficient memory 5244 of the
signal processing device shown in FIG. 359 with reference to FIG.
361. Note that the learning unit 5252 shown in FIG. 361 is
essentially the same as the learning unit 5152 shown in FIG. 341,
image memory 5261, a class tap extracting unit 5262, features
extracting unit 5263, class classification unit 5264, prediction
tap extracting unit 5265, supplementing computing unit 5266,
learning memory 5267, normal equation computing unit 5268, and
coefficient memory 5254 of the learning unit 5252 are the same as
the image memory 5161, class tap extracting unit 5162, features
extracting unit 5163, class classification unit 5164, prediction
tap extracting unit 5165, supplementing computing unit 5166,
learning memory 5167, normal equation computing unit 5168, and
coefficient memory 5154 of the learning unit 5152, so description
thereof will be omitted.
Also, as shown in FIG. 362, a 1/16 average processing unit 5253a of
a tutor image generating unit 5253, and an OLPF simulation
processing unit 5251a of a student image generating unit 5251 are
the same as the 1/16 average processing unit 5153a of the tutor
image generating unit 5153, and the OLPF simulation processing unit
5151a of a student image generating unit 5251 shown in FIG. 344, so
description thereof will be omitted as well.
The 1/64 average processing unit 5251b of the student image
generating unit 5251 regards each pixel of the HD image, which was
subjected to processing by the OLPF 5103 with the OLPF simulation,
serving as an input image as the light cast upon the sensor 2, and
regards the range of 8 pixels.times.8 pixels of the HD image as a
single pixel of the SD image, thereby generating a kind of spatial
integration effects, and virtually generating an image (SD image
without OLPF), which is to be generated at the sensor 2, with no
influence due to the OLPF 5103.
Next, description will be made regarding the learning processing by
the learning device shown in FIG. 361 with reference to the
flowchart shown in FIG. 363.
Note that the processing of step S5231 and the processing of steps
S5233 through S5241 are the same as the processing of step S5031
and the processing of steps S5033 through S5041, which have been
described with reference to the flowchart shown in FIG. 349, so
description thereof will be omitted.
In step S5232, the 1/64 average processing unit 5251b obtains an
average pixel value in increments of 64 pixels in total of 8
pixels.times.8 pixels regarding the image subjected to the OLPF
simulation processing input from the OLPF simulation processing
unit 5251a, further replaces the pixel values of the 64 pixels with
the average pixel value thereof in order, generates a student
image, which becomes an SD image in appearance, and outputs this to
the image memory 5261 of the learning unit 5252.
According to the above processing, prediction coefficients in the
case of taking an HD image without OLPF as a tutor image, and
taking an SD image with OLPF as a student image are to be stored in
the coefficient memory 5254. Further, copying the prediction
coefficients stored in this coefficient memory 5254 into the
coefficient memory 5244 of the signal processing device 5221, or
the like enables the signal processing shown in FIG. 360 to be
executed, and further, enables an SD image with OLPF to be
converted into an HD image without OLPF.
Summarizing the above processing, an actual world image is
subjected to the OLPF processing, and further, an SD image (actual
world+LPF+imaging device in the drawing) picked up by the imaging
device (sensor 2) is converted into an SD image (actual
world+imaging device in the drawing) from which the processing by
the OLPF is removed by the OLPF removing unit 5131 shown in FIG.
337, such as shown in the arrow A in FIG. 364, and further, the
actual world prior to the processing by the OLPF is estimated by
the continuity detecting unit 101 and the actual world estimating
unit 102, such as shown in the arrow A' in FIG. 364.
Also, the actual world estimating unit 102 shown in FIG. 352
estimates the actual world prior to the processing by the OLPF from
an SD image (actual world+LPF+imaging device in the drawing), such
as shown in the arrow B in FIG. 364.
Further, the signal processing device 5221 shown in FIG. 359
generates an HD image wherein the actual world is picked up by the
imaging device in a state without the influence by the OLPF from an
SD image (actual world+LPF+imaging device in the drawing), such as
shown in the arrow C in FIG. 364.
Also, the conventional class classification adaptation processing
generates an HD image wherein the actual world is picked up by the
imaging device in a state via the OLPF from an SD image (actual
world+LPF+imaging device in the drawing), such as shown in the
arrow D in FIG. 364.
Further, the signal processing device shown in FIG. 3 estimates the
actual world influenced by the OLPF from an SD image (actual
world+LPF+imaging device in the drawing), such as shown in the
arrow E in FIG. 364.
According to the above arrangement, image data corresponding to the
light signals when the light signals corresponding to the second
image data passes through the optical low pass filter is computed,
this is output as first image data, the multiple pixels
corresponding to the pixel of interest within the second image data
are extracted from the first image data, and learning is made so as
to predict the pixel value of the pixel of interest from the pixel
values of the extracted multiple pixels, thereby enabling an image
faithful as to the actual world to be generated.
Also, with the above arrangement, the approximation function f(x),
which approximates the actual world, has been handled as a
continuous function, but for example, the approximation function
f(x) may be set discontinuously for each region.
That is to say, the above arrangement has been made wherein the
function (approximation function) of the curve (curve shown with a
dotted line in the drawing) serving as a one-dimensional
cross-section indicating actual world light intensity distribution
is approximated with a polynomial, such as shown in FIG. 365, and
the actual world is estimated utilizing that this curve
continuously exists in the continuity direction.
However, this curve serving as a cross-section need not always to
be a continuous function such as a polynomial, for example, this
may be a discontinuous function, which varies for each region, such
as shown in FIG. 366. That is to say, in the case of FIG. 366, when
a region is a.sub.1.ltoreq.x<a.sub.2, the approximation function
f(x)=w.sub.1, when a region is a.sub.2.ltoreq.x<a.sub.3, the
approximation function f(x)=w.sub.2, when a region is
a.sub.3.ltoreq.x<a.sub.4, the approximation function
f(x)=w.sub.3, when a region is a.sub.4.ltoreq.x<a.sub.5, the
approximation function f(x)=w.sub.4, and further, when a region is
a.sub.5.ltoreq.x<a.sub.6, the approximation function
f(x)=w.sub.5, thus the different approximation function f(x) is set
for each region. Also, it can be conceived that w.sub.i is
essentially a level of the light intensity for each region.
Thus, the discontinuous function such as shown in FIG. 366 is
defined as in the following Expression (271) serving as a general
expression. f(x)=w.sub.i(a.sub.i.ltoreq.x<a.sub.i+1) (271)
Here, i represents the number of regions which are set.
Thus, a cross-sectional distribution (corresponding to a
cross-sectional curve) such as shown in FIG. 366 is set as a
constant for each region. Note that the cross-sectional
distribution of pixel values shown in FIG. 366 is extremely
different from the distribution of the curve shown with a dotted
line in FIG. 365 regarding the shape thereof, but actually, it
becomes possible to set a level which can approximate the
cross-sectional distribution of a discontinuous function with the
cross-sectional curve of a continuous function geometrically by
reducing the width of a range (in this case,
a.sub.i.ltoreq.x<a.sub.i+1) wherein each function f(x) is set,
to a minute width.
Accordingly, a pixel value P can be obtained with the following
Expression (272) by employing the approximation function f(x) made
up of an actual world discontinuous function, which is defined as
in Expression (271).
.intg..times..function..times..times.d ##EQU00166##
Here, X.sub.e and X.sub.s represent an integral range in the X
direction, wherein X.sub.s represents an integration start
position, and X.sub.e represents an integration end position
respectively.
However, it is actually difficult to directly obtain a function,
which approximates the actual world, such as shown in the above
Expression (271).
We can assume that the cross-sectional distribution of pixel values
such as shown in FIG. 366 continuously exists as to the continuity
direction, so that the distribution of the light intensity in the
space becomes like that shown in FIG. 367. The left portion of FIG.
367 corresponds to the distribution of pixel values in the case in
which the approximation function f(x) made up of a continuous
function continuously exists in the continuity direction, and the
right portion of FIG. 367, which is the same distribution
corresponding to the left portion, corresponds to the distribution
of pixel values in the case in which the approximation function
f(x) made up of a discontinuous function continuously exists in the
continuity direction.
That is to say, a state in which the cross-sectional shape shown in
FIG. 366 continues in the continuity direction is provided, so in
the event of employing the approximation function f(x) made up of a
discontinuous function, each level w.sub.i distributes in a band
shape in the continuity direction.
In order to determine the level of each region using the
approximation function f(x) defined by a discontinuous function
such as shown in the right portion of FIG. 367, it is necessary to
obtain the sum of products between the weight according to the
proportion of the area for each region occupied in a range wherein
each level (each function) is set, of the total area of the pixels,
and the level thereof, generate a normal equation using the pixel
value of the corresponding pixel, and obtain the pixel value of
each region using the least square method.
That is to say, as shown in FIG. 368, in the event that a
discontinuous function distributes such as shown in the left
portion of FIG. 368, in the case of obtaining the pixel value of
the pixel of interest (note that FIG. 368 is a top view
illustrating a pixel array when taking the paper space as an X-Y
plane, and each grid corresponds to a pixel) shown in a grid
surrounded with a thick line in FIG. 368, a triangular (triangle of
which the bottom side is up) range present above the hatched
portion of the pixel of interest is a range set by f(x)=W.sub.2,
the hatched portion is a range set by f(x)=W.sub.3, and the
triangular (triangle of which the bottom side is down) range
present below the hatched portion is a range set by
f(x)=W.sub.4.
In the event that the area of the pixel of interest is 1, if we say
that the proportion occupied by the range by f(x)=W.sub.2 is 0.2,
the proportion occupied by the range by f(x)=W.sub.3 is 0.5, and
the proportion occupied by the range by f(x)=W.sub.4 is 0.3, the
pixel value P of the pixel of interest is represented with the sum
of products of the pixel value and proportion for each range, so is
obtained by computation shown in the following Expression (273).
P=0.2.times.W.sub.2+0.5.times.W.sub.3+0.3.times.W.sub.4 (273)
Accordingly, the levels of pixel values can be obtained by
generating a relational expression as to pixels regarding each
pixel using the relationship shown in Expression (273), for
example, in order to obtain the levels w.sub.1 through w.sub.5, if
Expression (273) indicating the relationship with the pixel values
of at least five pixels including all of the levels can be
obtained, it becomes possible to obtain w.sub.1 through w.sub.5
indicating the levels of the pixel values using the least square
method (simultaneous equations in the event that the number of
relational expressions are the same as the number of unknowns).
Thus, it becomes possible to obtain the approximation function f(x)
made up of a discontinuous function by employing the
two-dimensional relationship with continuity.
Also, since the angle .theta. as continuity is determined by the
continuity detecting unit 101, the straight line having the angle
.theta. passing through the origin (0, 0) is uniquely determined,
and a position x.sub.1 in the X direction of the straight line at
an arbitrary position y in the Y direction is represented as the
following Expression (274). However, in Expression (274), s
represents a gradient as continuity, which is represented with cot
.theta. (=s) when the gradient as continuity is represented with
the angle .theta.. x.sub.1=s.times.y (274)
That is to say, a point on the straight line corresponding to
continuity of data is represented with a coordinate value (x.sub.1,
y).
According to Expression (274), a cross-sectional direction distance
x' (distance shifted in the X direction along the straight line
wherein continuity exists) is represented as in the following
Expression (275). x'=x-x.sub.1=x-s.times.y (275)
Accordingly, the approximation function f(x, y) at an arbitrary
position (x, y) is represented as the following Expression (276)
using Expression (271) and Expression (275).
f(x,y)=w.sub.i(a.sub.i.ltoreq.(x-s.times.y)<a.sub.i+1) (276)
Note that in Expression (276), it can be said that w.sub.i is
features indicating the light intensity level in each region.
Hereafter, w.sub.i is also referred to as features.
Accordingly, the actual world estimating unit 102 can estimate a
waveform F(x, y) by estimating the approximation function f(x, y)
made up of a discontinuous function as long as the features w.sub.i
for each region of Expression (276) can be computed.
Consequently, hereafter, description will be made regarding a
method for computing the features w.sub.i of Expression (276).
That is to say, upon the approximation function f(x, y) represented
with Expression (276) being integrated with an integral range
(integral range in the spatial direction) corresponding to a pixel
(the detecting element of the sensor 2), the integral value becomes
the estimated value regarding the pixel value of the pixel. It is
the following Expression (277) that this is represented with an
equation. Note that with the two-dimensional polynomial
approximating method employing a discontinuous function, the frame
direction T is regarded as a constant value, so Expression (277) is
taken as an equation of which variables are the positions x and y
in the spatial directions (X direction and Y direction).
.function..intg..times..intg..times..function..times..times.d.times..time-
s.d ##EQU00167##
In Expression (277), P (x, y) represents the pixel value of a pixel
of which the center position is in a position (x, y) (relative
position (x, y) from the pixel of interest) of an input image from
the sensor 2.
Thus, with the two-dimensional approximating method, the
relationship between the input pixel value P(x, y) and the
two-dimensional approximation function f(x, y) can be represented
with Expression (277), and accordingly, the actual world estimating
unit 102 can estimate the two-dimensional function F(x, y)
(waveform F(x, y) wherein the light signal in the actual world 1
having continuity in the spatial directions) by computing the
features w.sub.i with, for example, by the least square method or
the like using Expression (277).
Now, description will be made regarding the configuration of the
actual world estimating unit 102, which sets the approximation
function f(x) using a discontinuous function as described above,
and estimates the actual world, with reference to FIG. 369.
As shown in FIG. 369, the actual world estimating unit 102 includes
a condition setting unit 5301, input image storing unit 5302, input
pixel value acquiring unit 5303, integration component computing
unit 5304, normal equation generating unit 5305, and approximation
function generating unit 5306.
The condition setting unit 5301 sets a pixel range (tap range) used
for estimating the function F(x, y) corresponding to a pixel of
interest, and a range (e.g., width of
a.sub.i.ltoreq.x<a.sub.i+1, the number of i) of the
approximation function f(x, y).
The input image storing unit 5302 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel value acquiring unit 5303 acquires, of the input
images stored in the input image storing unit 5302, an input image
region corresponding to the tap range set by the condition setting
unit 5301, and supplies this to the normal equation generating unit
5305 as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described. Note that
a specific example of the input pixel value table will be described
later.
Incidentally, as described above, the actual world estimating unit
102 employing the two-dimensional approximating method computes the
features wi of the approximation function f(x, y) represented with
the above Expression (276) by solving the above Expression (277)
using the least square method.
Expression (277) can be represented as in the following Expression
(278).
.function..times..times..function. ##EQU00168##
In Expression (278), T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e)
represents the integration results of a region serving as features
wi (region serving as the light level wi), of regions serving as
integral ranges, i.e., represents an area. Hereafter,
T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e) is referred to as
integral components.
The integration component computing unit 5304 computes the integral
components T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e) (=(x-0.5,
x+0.5, y-0.5, y+0.5): in the case of obtaining a region for the
worth of a single pixel).
Specifically, the integral components T.sub.i(x.sub.s, x.sub.e,
y.sub.s, y.sub.e) shown in Expression (278) are for obtaining the
area of certain features w.sub.i, of the pixels to be obtained, as
described with reference to FIG. 368. Accordingly, the integration
component computing unit 5304 may obtain T.sub.i(x.sub.s, x.sub.e,
y.sub.s, y.sub.e) by obtaining an area occupied for each features
w.sub.i geometrically based on the width d for each features and
the angle .theta. information of data continuity, or by performing
multiple division and integration according to the Simpson's rule,
rather, a method for obtaining an area is not restricted to those,
for example, an area may be obtained by the Monte Carlo Method.
As described in FIG. 368, the features wi can be computed as long
as the width of a.sub.i.ltoreq.(x-s.times.y)<a.sub.i+1, a
variable s indicating the gradient of continuity, and the relative
pixel positions (x, y) are known. Of these, the relative pixel
positions (x, y) are determined with a pixel of interest, and a tap
range, the variable s is cot .theta., which is determined with the
angle .theta., and the width of
a.sub.i.ltoreq.(x-s.times.y)<a.sub.i+1 is set beforehand, and
accordingly, each value becomes a known value.
Accordingly, the integration component computing unit 5304 computes
the integral components T.sub.i (x-0.5, x+0.5, y-0.5, y+0.5) based
on the tap range and the width set by the condition setting unit
5301, and the angle .theta. of the data continuity information
output from the data continuity detecting unit 101, and supplies
the computed results to the normal equation generating unit 5305 as
an integration component table.
The normal equation generating unit 5305 generates a normal
equation in the case of obtaining the above Expression (277), i.e.,
Expression (278) by the least square method using the input pixel
value table supplied from the input pixel value acquiring unit
5303, and the integration component table supplied from the
integration component computing unit 5304, and outputs this to the
approximation function generating unit 5306 as a normal equation
table. Note that a specific example of a normal equation will be
described later.
The approximation function generating unit 5306 computes the
respective features w.sub.i of the above Expression (278) by
solving the normal equation included in the normal equation table
supplied from the normal equation generating unit 5305 using the
matrix solution, and output these to the image generating unit
103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40) to which
the two-dimensional approximating method employing a discontinuous
function is applied, with reference to the flowchart in FIG.
370.
For example, let us say that the light signal in the actual world 1
having continuity in the spatial direction represented with the
gradient G.sub.F has been detected by the sensor 2, and has been
stored in the input image storing unit 5302 as an input image
corresponding to one frame. Also, let us say that the data
continuity detecting unit 101 has output the angle .theta. in the
continuity detecting processing in step S101 (FIG. 406) as data
continuity information of the input image.
In this case, in step S5301, the condition setting unit 5301 sets
conditions (a tap range, the width of a.sub.i.ltoreq.x<a.sub.i+1
(the width of the same features), and the number of i).
For example, let us say that a tap range shown in FIG. 371 has been
set, and also d has been set as the width.
FIG. 371 is a diagram for describing an example of a tap range. In
FIG. 371, the X direction and Y direction represent the X direction
and Y direction of the sensor 2. Also, the tap range represents a
pixel group made up of 15 pixels (15 grids surrounded with a thick
line on the right portion in the drawing) in total of the right
portion in FIG. 371.
Further, as shown in FIG. 371, let us say that a pixel of interest
has been set to a pixel of the hatched portion in the drawing, of
the tap range. Also, let us say that each pixel is denoted with a
number l such as shown in FIG. 371 (l is any integer value of 0
through 14) according to the relative pixel positions (x, y) from
the pixel of interest (a coordinate value of a pixel-of-interest
coordinates system wherein the center (0, 0) of the pixel of
interest is taken as the origin).
Now, description will return to FIG. 370, wherein in step S5302,
the condition setting unit 5301 sets a pixel of interest.
In step S5303, the input pixel value acquiring unit 5303 acquires
an input pixel value based on the condition (tap range) set by the
condition setting unit 5301, and generates an input pixel value
table. That is to say, in this case, the input pixel value
acquiring unit 5303 acquires the pixel values of the pixels of the
input image region (pixels appended with numbers 0 through 14 in
FIG. 371), generates a table made up of 15 input pixel values P (l)
as an input pixel value table.
In step S5304, the integration component computing unit 5304
computes integral components based on the conditions (a tap range,
width, the number of i) set by the condition setting unit 5301, and
the data continuity information (angle .theta.) supplied from the
data continuity detecting unit 101, and generates an integration
component table.
In this case, the integration component computing unit 5304
computes the integral components T.sub.i(x.sub.s, x.sub.e, y.sub.s,
y.sub.e) (=T.sub.i (x-0.5, x+0.5, y-0.5, y+0.5): in the case of
expressing one pixel size as 1.times.1) in the above Expression
(278) as a function of l such as the integral components T.sub.i
(l) shown in the left side of the following Expression (279).
T.sub.i(l)=T.sub.i(x-0.5,x+0.5,y-0.5,y+0.5) (279)
That is to say, in this case, if we say that i is 0 through 5, the
90 T.sub.i (l) in total of the 15 T.sub.0 (l), 15 T.sub.1 (l), 15
T.sub.2 (l), 15 T.sub.3 (l), 15 T.sub.4 (l), and 15 T.sub.5 (l) are
computed, and an integration component table including these is
generated.
Note that the sequence of the processing in step S5303 and the
processing in step S5304 is not restricted to the example in FIG.
370, the processing in step S5304 may be executed first, or the
processing in step S5303 and the processing in step S5304 may be
executed simultaneously.
Next, in step S5305, the normal equation generating unit 5305
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 5303 at the
processing in step S5303, and the integration component table
generated by the integration component computing unit 5304 at the
processing in step S5304.
Specifically, in this case, the features w.sub.i are computed with
the least square method using the above Expression (278), so a
normal equation corresponding thereto is represented as in the
following Expression (280).
.times..times..function..times..function..times..times..function..times..-
function..times..times..function..times..function..times..times..function.-
.times..function..times..times..function..times..function..times..times..f-
unction..times..function..times..times..function..times..function..times..-
times..function..times..function..times..times..function..times..function.-
.times..times..times..function..times..function..times..times..function..t-
imes..function..times..times..function..times..function.
##EQU00169##
Note that in Expression (280), L represents the maximum value of
the pixel number l in the tap range. n represents the number of i
of the features w.sub.i which defines the approximation function
f(x). Specifically, in this case, L=15.
If we define each matrix of the normal equation shown in Expression
(280) as the following Expressions (281) through (283), the normal
equation is represented as in the following Expression (284).
.times..times..function..times..function..times..times..function..times..-
function..times..times..function..times..function..times..times..function.-
.times..function..times..times..function..times..function..times..times..f-
unction..times..function..times..times..function..times..function..times..-
times..function..times..function..times..times..function..times..function.-
.times..times..function..times..function..times..times..function..times..f-
unction..times..times..function..times..function..times.
##EQU00170##
As shown in Expression (282), the respective components of the
matrix W.sub.MAT are the features w.sub.i to be obtained.
Accordingly, in Expression (284), if the matrix T.sub.MAT of the
left side and the matrix P.sub.MAT of the right side are
determined, the matrix W.sub.MAT may be computed with the matrix
solution.
Specifically, as shown in Expression (281), the respective
components of the matrix T.sub.MAT may be calculated with the above
integral components T.sub.i (l). That is to say, the integral
components T.sub.i (l) are included in the integration component
table supplied from the integration component computing unit 5304,
so the normal equation generating unit 5305 can calculate each
component of the matrix T.sub.MAT using the integration component
table.
Also, as shown in Expression (283), the respective components of
the matrix P.sub.MAT may be computed with the integral components
T.sub.i (l) and the input pixel values P (l). That is to say, the
integral components T.sub.i (l) is the same as those included in
the respective components of the matrix T.sub.MAT, also the input
pixel values P (l) are included in the input pixel value table
supplied from the input pixel value acquiring unit 5303, so the
normal equation generating unit 5305 can calculate each component
of the matrix P.sub.MAT using the integration component table and
input pixel value table.
Thus, the normal equation generating unit 5305 computes each
component of the matrix T.sub.MAT and matrix P.sub.MAT, and outputs
the computed results (each component of the matrix T.sub.MAT and
matrix P.sub.MAT) to the approximation function generating unit
5306 as a normal equation table.
Upon the normal equation table being output from the normal
equation generating unit 5305, in step S5306, the approximation
function generating unit 5306 computes the features w.sub.i (i.e.,
the levels w.sub.i, which are defined for each region, of the
two-dimensional approximation function f(x, y) made up of a
discontinuous function) serving as the respective components of the
matrix W.sub.MAT in the above Expression (284) based on the normal
equation table.
Specifically, the normal equation in the above Expression (284) can
be transformed as the following Expression (285).
.times. ##EQU00171##
In Expression (285), the respective components of the matrix
W.sub.MAT in the left side are the features w.sub.i to be obtained.
The respective components regarding the matrix T.sub.MAT and matrix
P.sub.MAT are included in the normal equation table supplied from
the normal equation generating unit 5305. Accordingly, the
approximation function generating unit 5306 computes the matrix
W.sub.MAT by computing the matrix in the right side of Expression
(285) using the normal equation table, and outputs the computed
results (features w.sub.i) to the image generating unit 103.
In step S5307, the approximation function generating unit 5306
determines regarding whether or not the processing of all the
pixels has been completed.
In step S5307, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S5302, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S5302 through S5307 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S5307, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
As description of the two-dimensional approximating method
employing a discontinuous function, an example for calculating the
features w.sub.i of the approximation function f(x, y)
corresponding to the spatial directions (X direction and Y
direction) has been employed, but the two-dimensional approximating
method employing a discontinuous function can be applied to the
temporal and spatial directions (X direction and T direction, or Y
direction and T direction) as well.
That is to say, the above example is an example in the case of the
light signal in the actual world 1 having continuity in the spatial
direction, and accordingly, the equation including two-dimensional
integration in the spatial directions (X direction and Y
direction), such as shown in the above Expression (277). However,
the concept regarding two-dimensional integration can be applied
not only to the spatial direction but also to the time-space
directions (X direction and T direction, or Y direction and T
direction).
In other words, with the two-dimensional approximating method
employing a discontinuous function, even in the case in which the
light signal function F(x, y, t), which needs to be estimated, has
not only continuity in the spatial direction but also continuity in
the time-space directions (however, X direction and T direction, or
Y direction and T direction), this can be approximated with a
two-dimensional discontinuous function.
Specifically, for example, in the event that an object (toy plane
in the drawing) D1 (image in the bottom frame in the drawing) such
as shown in FIG. 372 moves horizontally in the X direction at
uniform velocity to an object D2 (image in the middle frame in the
drawing), movement of the object is represented with like a track
L1 in the X-T plane such as shown in the upper portion of FIG. 372.
Note that the upper portion of FIG. 372 illustrates change in the
pixel value on the surface wherein OPQR in the drawing are taken as
apexes.
In other words, it can be said that the track L1 represents the
direction of continuity in the time-space directions in the X-T
plane. Accordingly, the data continuity detecting unit 101 can
output a traced angle such as shown in FIG. 372 (strictly speaking,
though not shown in the drawing, an angle between the direction of
data continuity serving as a tack (the above movement) when the
object moves from D1 to D2 and the X direction in the spatial
direction) as data continuity information corresponding to the
gradient (angle as continuity) representing continuity in the
time-space directions in the X-T plane as well as the above angle
.theta. (data continuity information corresponding to continuity in
the spatial directions represented with a certain gradient (angle)
in the X-Y plane).
Accordingly, the actual world estimating unit 102 employing the
approximation technique using the two-dimensional discontinuous
function can compute the features w.sub.i of an approximation
function f(x, t) in the same method as the above method by
employing the movement .theta. instead of the angle .theta..
However, in this case, the equation to be employed is not the above
Expression (277) but the following Expression (286).
.function..intg..times..intg..times..function..times..times.d.times..time-
s.d ##EQU00172##
In the event of the processing on the X-T plane, the relationship
between each pixel and the discontinuous function shown in the
right portion of FIG. 371 becomes like that shown in FIG. 373. That
is to say, in FIG. 373, the cross-sectional shape in the spatial
direction X (cross-sectional shape with the discontinuous function)
continues in a certain continuity direction as to the frame
direction T. Consequently, in the event that the levels are five
types of levels w.sub.1 through W.sub.5, the band, which becomes
the same level as shown in the left portion of FIG. 371, is
distributed in the continuity direction.
Accordingly, in this case, a pixel value can be obtained by
employing a pixel present on the X-T plane such as shown in the
right portion of FIG. 373. Note that in the right portion of FIG.
373, each grid represents a pixel, and the X direction represents
the width of a pixel, but with regard to the frame direction, each
grid increment is equivalent to one frame.
Also, an approximation function f(y, t) focusing attention on the
spatial direction Y instead of the spatial direction X can be
handled in the same way as the above approximation function f(x,
t).
Description has been made regarding a method for setting a
two-dimensional approximation function made up of a discontinuous
function, and estimating the actual world so far, but further, a
three-dimensional approximation function made up of a discontinuous
function enables the actual world to be estimated as well.
For example, let us consider a two-dimensional discontinuous
function, which is different for each region, such as shown in FIG.
374. That is to say, in the case of FIG. 374, when a region is
a.sub.1.ltoreq.x<a.sub.2, and also b.sub.1.ltoreq.y<b.sub.2,
the approximation function is f(x, y)=w.sub.1, when a region is
a.sub.2.ltoreq.x<a.sub.3, and also b.sub.3.ltoreq.y<b.sub.4,
the approximation function is f(x, y)=w.sub.2, when a region is
a.sub.3.ltoreq.x<a.sub.4, and also b.sub.5.ltoreq.y<b.sub.6,
the approximation function is f(x, y)=w.sub.3, when a region is
a.sub.4.ltoreq.x<a.sub.5, and also b.sub.7.ltoreq.y<b.sub.8,
the approximation function is f(x, y)=w.sub.4, and further, when a
region is a.sub.3.ltoreq.x<a.sub.4, and also
b.sub.9.ltoreq.y<b.sub.10, the approximation function is f(x,
y)=w.sub.5, thus the different approximation function f(x, y) is
set for each region. Also, it can be conceived that w.sub.i is
essentially a level of the light intensity for each region.
Thus, the discontinuous function such as shown in FIG. 374 is
defined as the following Expression (287) serving as a general
equation.
f(x,y)=wi(a.sub.j.ltoreq.x<a.sub.j+1&b.sub.2k-1.ltoreq.y<b.sub.2k)
(287)
Note that j and k are arbitrary integers, but i is a sequential
number for identifying a region, which can be expressed by a
combination of j and k.
Thus, a cross-sectional distribution (corresponding to a
cross-sectional curve) such as shown in FIG. 374 is set as a
constant for each region.
Accordingly, a pixel value P(x, y) can be obtained with the
following Expression (288) by employing the approximation function
f(x, y) made up of a discontinuous function in the actual world,
which is defined as in Expression (287).
.function..intg..times..intg..times..function..times..times.d.times..time-
s.d ##EQU00173##
Here, x.sub.e and x.sub.s represent an integral range in the X
direction, wherein x.sub.s represents an integration start position
in the X direction, and x.sub.e represents an integration end
position in the X direction respectively. Similarly, y.sub.e and
y.sub.s represent an integral range in the Y direction, wherein
y.sub.s represents an integration start position in the Y
direction, and y.sub.e represents an integration end position in
the Y direction respectively.
However, it is difficult to directly obtain a function which
approximates the actual world, such as shown in the above
Expression (287), in actual practice.
We can assume that the cross-sectional distribution of pixel values
such as shown in FIG. 374 continuously exists as to the continuity
direction in the frame direction, so that the distribution of the
light intensity in the space becomes like that shown in FIG. 375.
The left portion of FIG. 375 illustrates the distribution of pixel
values on the X-T plane in the case in which the approximation
function f(x, y) made up of a discontinuous function continuously
exists in the continuity direction of the frame direction and the X
direction, and the right portion of FIG. 374 illustrates a
distribution wherein the cross-section of the light intensity level
on the X-Y plane continues in the frame direction.
That is to say, a state in which the cross-sectional shape shown in
FIG. 374 continues in the continuity direction is provided, so the
region of each level w.sub.i distributes in a rod shape in the
continuity direction such as shown in the right portion of FIG.
375.
In order to determine the pixel value of each three-dimensional
region using the approximation function f(x, y) defined by a
discontinuous function such as shown in the right portion of FIG.
375, with the above two-dimensions, a proportion according to a
volume is employed for computation, as with the method employing an
area. That is to say, of the total volume (three-dimensional
volumes made up of the X direction, Y direction, and T direction)
of each pixel, the sum of products of weight according to the
proportion of volumes occupied by a range wherein each level is
set, and the level thereof is obtained, the pixel value of the
corresponding pixel is employed, thereby obtaining the pixel value
of each region with the least square method.
That is to say, as shown in FIG. 376, let us say that the level of
one region is f(x, y)=w.sub.1, and the level of the other region is
f(x, y)=w.sub.2, with a boundary R as a boundary. Also, let us say
that a cube made up of ABCDEFGH in the drawing in the XYT space
represents a pixel of interest. Further, let us say that the
cross-section with the boundary R in the pixel of interest is a
rectangle made up of IJKL.
Also, let us say that of the volume of the pixel P, the proportion
occupied by a portion serving as a triangle pole made up of IBJ-KFL
is represented with M1, and the proportion occupied by the volumes
of the portions other than that (pentangular pole made up of
ADCJI-EGHLK) is represented with M2. Note that the term "volume"
here means represents the magnitude of an occupied region on the
XYt space.
At this time, the pixel value P of the pixel of interest is
represented with the sum of products of the pixel value of each
range and the proportion, and accordingly, can be obtained by the
computation shown in the following Expression (289).
P=M1.times.w.sub.1+M2.times.w.sub.2 (289)
Accordingly, the levels of pixel values can be obtained by
generating an expression indicating the relationship as to pixel
values regarding each pixel using the relationship shown in
Expression (289), for example, in order to obtain w.sub.1 through
w.sub.2 as coefficients indicating pixel values, if Expression
(289) indicating the relationship with the pixel values of at least
two pixels including each coefficient can be obtained, it becomes
possible to obtain w.sub.1 through w.sub.2 indicating the levels of
the pixel values using the least square method (simultaneous
equations in the event that the number of relational expressions
are the same as the number of unknowns).
Thus, it becomes possible to obtain the approximation function f(x,
y) made up of a discontinuous function by employing the
three-dimensional relationship with continuity.
For example, velocities v.sub.x and v.sub.y (essentially, the
gradients of the X-T plane and Y-T plane) on the X-T plane and in
an Y-T plane shape can be obtained based on the movement .theta.
which is equivalent to the angle .theta. as continuity in an X-Y
planar shape output from the continuity detecting unit 101, and
accordingly, a position x.sub.1 in the X direction and a position
y.sub.1 in the X direction of the straight line of continuity at an
arbitrary position (x, y) in the X direction and Y direction are
represented as in the following Expression (290).
x.sub.1=v.sub.x.times.t,y.sub.1=v.sub.y.times.t (290)
That is to say, a point on the straight line corresponding to
continuity of data is represented with a coordinate value (x.sub.1,
y.sub.1).
According to Expression (290), cross-sectional direction distances
x' and y' (shifted distances in the X direction and Y direction
along the straight line where continuity exists) are represented as
in the following Expression (291). x'=x-x.sub.1=x-v.sub.x.times.t
y'=y-y.sub.1=y-v.sub.y.times.t (291)
Accordingly, the approximation function f(x, y) at an arbitrary
position (x, y) in the input image is represented as in the
following Expression (292) according to Expression (287) and
Expression (291).
f(x,y,t)=w.sub.i(a.sub.j.ltoreq.(x-v.sub.x.times.t)<a.sub.j+1&b.sub.2k-
-1.ltoreq.(y-v.sub.y.times.t)<b.sub.2k) (292)
Accordingly, if the actual world estimating unit 102 can compute
the features w.sub.i for each region of Expression (292), the
actual world estimating unit 102 can estimate a waveform F(x, y, t)
by estimating an approximation function f(x, y, t) made up of a
discontinuous function.
Consequently, hereafter, description will be made regarding a
method for computing the features w.sub.i of Expression (292).
That is to say, upon the approximation function f(x, y, t)
represented with Expression (292) being subjected to integration
with an integral range (integral range in the spatial direction)
corresponding to a pixel (the detecting element of the sensor 2),
the integral value becomes the estimated value regarding the pixel
value of the pixel. It is the following Expression (293) that this
is represented with an equation.
.function..intg..times..intg..times..intg..times..function..times..times.-
d.times..times.d.times..times.d ##EQU00174##
In Expression (293), P (x, y, t) represents the pixel value of a
pixel of which the center position is in a position (x, y, t)
(relative position (x, y, t) from the pixel of interest) of an
input image from the sensor 2.
Thus, with the three-dimensional approximating method, the
relationship between the input pixel value P (x, y, t) and the
three-dimensional approximation function f(x, y, t) made up of a
discontinuous function can be represented with Expression (293),
and accordingly, the actual world estimating unit 102 can estimate
the three-dimensional function F(x, y, t) (waveform F(x, y, t)
wherein the light signal in the actual world 1 having continuity in
the spatial direction is represented focusing attention on the
time-space directions) by computing the features w.sub.i with, for
example, the least square method or the like using Expression
(293).
Next, description will be made regarding the configuration of the
actual world estimating unit 102, which sets the three-dimensional
approximation function f(x, y, t) made up of a discontinuous
function as described above, and estimates the actual world, with
reference to FIG. 377.
As shown in FIG. 377, the actual world estimating unit 102 includes
a condition setting unit 5321, input image storing unit 5322, input
pixel value acquiring unit 5323, integration component computing
unit 5304, normal equation generating unit 5325, and approximation
function generating unit 5326.
The condition setting unit 5321 sets a pixel range (tap range) used
for estimating the function F(x, y, t) corresponding to a pixel of
interest, and the range (e.g., width of
a.sub.j.ltoreq.(x-v.sub.x.times.t)<a.sub.j+1 &
b.sub.2k-1.ltoreq.(y-v.sub.y.times.t)<b.sub.2k, the number of i)
of the approximation function f(x, y, t).
The input image storing unit 5322 temporarily stores an input image
(pixel values) from the sensor 2.
The input pixel value acquiring unit 5323 acquires, of the input
images stored in the input image storage unit 5322, an input image
region corresponding to the tap range set by the condition setting
unit 5321, and supplies this to the normal equation generating unit
5325 as an input pixel value table. That is to say, the input pixel
value table is a table in which the respective pixel values of
pixels included in the input image region are described. Note that
a specific example of the input pixel value table will be described
later.
Incidentally, as described above, the actual world estimating unit
102 employing the three-dimensional approximating method computes
the features w.sub.i of the approximation function f(x, y, t)
represented with the above Expression (292) by solving the above
Expression (293) using the least square method.
Expression (293) can be represented as in the following Expression
(294).
.function..times..times..function. ##EQU00175##
In Expression (294), T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e,
t.sub.s, t.sub.e) represents, of the regions serving as an integral
range, the integration result of the region serving as the features
w.sub.i (region serving as the light levels w.sub.i), i.e.,
volumes. Hereafter, T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e,
t.sub.s, t.sub.e) is referred to as integral components. Note that
this Expression (294) corresponds to the integral components
T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e) in a two-dimensional
arithmetic operation.
The integration component computing unit 5324 computes the integral
components T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e, t.sub.s,
t.sub.e) (=(x-0.5, x+0.5, y-0.5, y+0.5, t-0.5, t+0.5): in the case
of acquiring one pixel worth region).
Specifically, the integral components T.sub.i(x.sub.s, x.sub.e,
y.sub.s, y.sub.e, t.sub.s, t.sub.e) shown in Expression (294) are
for obtaining the volumes of the predetermined features w.sub.i, of
the pixels to be obtained, as described with reference to FIG. 376.
Accordingly, the integration component computing unit 5324 may
obtain T.sub.i(x.sub.s, x.sub.e, y.sub.s, y.sub.e, t.sub.s,
t.sub.e) by obtaining volumes occupied for each features w.sub.i
geometrically based on the widths d and e for each features and the
continuity direction information (e.g., the angle .theta. as to a
certain axis of continuity), or by performing multiple division and
integration according to the Simpson's rule, rather, a method for
obtaining volumes is not restricted to those, for example, volumes
may be obtained by the Monte Carlo Method.
As described in FIG. 376, the features w.sub.i can be computed as
long as the width of
a.sub.j.ltoreq.(x-v.sub.x.times.t)<a.sub.j+1 &
b.sub.2k-1.ltoreq.(y-v.sub.y.times.t)<b.sub.2k, and the
continuity direction information (e.g., the velocities v.sub.x and
v.sub.y, or the angle .theta. as to a certain axis of continuity),
and the relative pixel positions (x, y, t) are known. Of these, the
relative pixel positions (x, y, t) are determined with a pixel of
interest, and a tap range, the continuity information is determined
with the information detected by the continuity detecting unit 101,
and the width of a.sub.j.ltoreq.(x-v.sub.x.times.t)<a.sub.j+1
& b.sub.2k-1.ltoreq.(y-v.sub.y.times.t)<b.sub.2k is set
beforehand, and accordingly, each value becomes a known value.
Accordingly, the integration component computing unit 5324 computes
the integral components T.sub.i (x-0.5, x+0.5, y-0.5, y+0.5, t-0.5,
t+0.5) based on the tap range and the width set by the condition
setting unit 5321, and the data continuity information output from
the data continuity detecting unit 101, and supplies the computed
results to the normal equation generating unit 5325 as an
integration component table.
The normal equation generating unit 5325 generates a normal
equation in the case of obtaining the above Expression (293), i.e.,
Expression (294) by the least square method using the input pixel
value table supplied from the input pixel value acquiring unit
5323, and the integration component table supplied from the
integration component computing unit 5324, and outputs this to the
approximation function generating unit 5326 as a normal equation
table.
The approximation function generating unit 5326 computes the
respective features w.sub.i of the above Expression (294) by
solving the normal equation included in the normal equation table
supplied from the normal equation generating unit 5325 using the
matrix solution, and output these to the image generating unit
103.
Next, description will be made regarding the actual world
estimating processing (processing in step S102 in FIG. 40) to which
the three-dimensional approximating method employing a
discontinuous function is applied, with reference to the flowchart
in FIG. 378.
For example, let us say that the light signal in the actual world 1
having continuity in the time-space directions represented with the
velocities V.sub.x and V.sub.y as to the X-t plane and Y-t plane
has been detected by the sensor 2, and has been stored in the input
image storing unit 5322 as an input image corresponding to one
frame. Also, let us say that the data continuity detecting unit 101
has obtained the velocities V.sub.x and V.sub.y as the data
continuity information of the input image in the continuity
detecting processing in step S101 (FIG. 406).
In this case, in step S5321, the condition setting unit 5321 sets
conditions (a tap range, the width of
a.sub.j.ltoreq.(x-v.sub.x.times.t)<a.sub.j+1 &
b.sub.2k-1.ltoreq.(y-v.sub.y.times.t)<b.sub.2k (the same
features (the widths d and e of regions which becomes the same
approximation function)), and the number of i).
For example, let us say that the tap range shown in FIG. 379 has
been set, and also width in the horizontal direction.times.width in
the vertical direction=d.times.e has been set as widths.
The set tap range is assumed to be that shown in FIG. 379, for
example. In FIG. 379, the X direction and Y direction represent the
X direction and Y direction of the sensor 2. Also, t represents a
frame number, and the tap range represents a pixel group made up of
27 pixels in total of pixels P0 through P26 serving as 9 pixels per
frame.times.3 frames as shown in the right portion of FIG. 379.
Further, as shown in FIG. 379, a pixel of interest is assumed to be
set to the pixel P13 on the center portion in the frame number t=n
in the drawing. Also, let us say that each pixel is denoted with a
number l such as shown in FIG. 379 (l is any integer value of P0
through P26) according to the relative pixel positions (x, y, t)
from the pixel of interest (a coordinate value of a
pixel-of-interest coordinates system wherein the center (0, 0, 0)
of the pixel of interest is taken as the origin).
Now, description will return to FIG. 378, wherein in step S5322,
the condition setting unit 5321 sets a pixel of interest.
In step S5323, the input pixel value acquiring unit 5323 acquires
an input pixel value based on the condition (tap range) set by the
condition setting unit 5321, and generates an input pixel value
table. That is to say, in this case, the input pixel value
acquiring unit 5323 acquires the pixel values of the pixels in the
input image region (pixels denoted with the numbers P0 through P26
in FIG. 379), and generates a table made up of 27 input pixel
values P (l) as an input pixel value table.
In step S5324, the integration component computing unit 5324
computes integral components based on the conditions (a tap range,
width, and the number of i) set by the condition setting unit 5321,
and the data continuity information supplied from the data
continuity detecting unit 101, and generates an integration
component table.
In this case, the integration component computing unit 5324
computes the integral components T.sub.i(x.sub.s, x.sub.e, y.sub.s,
y.sub.e, t.sub.s, t.sub.e) (=T.sub.i (x-0.5, x+0.5, y-0.5, y+0.5,
t-0.5, t+0.5): in the case of expressing one pixel size as X
direction.times.Y direction.times.frame direction
t=1.times.1.times.1) in the above Expression (294) as a function of
1 such as the integral components T.sub.i(l) shown in the left side
of the following Expression (295).
T.sub.i(l)=T.sub.i(x-0.5,x+0.5,y-0.5,y+0.5,t-0.5, t+0.5) (295)
That is to say, in this case, if i is assumed to be 0 through 5,
the 162 T.sub.i(l) in total of the 27 T.sub.0(l), 27 T.sub.1(l), 27
T.sub.2(l), 27 T.sub.3(l), 27 T.sub.4(l), and 27 T.sub.5(l) are
computed, and an integration component table including these is
generated.
Note that the sequence of the processing in step S5323 and the
processing in step S5324 is not restricted to the example in FIG.
378, the processing in step S5324 may be executed first, or the
processing in step S5323 and the processing in step S5324 may be
executed simultaneously.
Next, in step S5325, the normal equation generating unit 5325
generates a normal equation table based on the input pixel value
table generated by the input pixel value acquiring unit 5323 at the
processing in step S5323, and the integration component table
generated by the integration component computing unit 5324 at the
processing in step S5324.
Specifically, in this case, the features w.sub.i are computed with
the least square method using the above Expression (295), so a
normal equation corresponding to this is represented as in the
following Expression (296).
.times..times..function..times..function..times..times..function..times..-
function..times..times..function..times..function..times..times..function.-
.times..function..times..times..function..times..function..times..times..f-
unction..times..function..times..times..function..times..function..times..-
times..function..times..function..times..times..function..times..function.-
.times..times..times..function..times..function..times..times..function..t-
imes..function..times..times..function..times..function.
##EQU00176##
Note that in Expression (296), L represents the maximum value of
the pixel number l in the tap range. n represents the number of i
of the features w.sub.i which defines the approximation function
f(x). v.sub.l represents weight. Specifically, in this case,
L=27.
This normal equation is the same format as the above Expression
(280), and employs the same technique as that in the above
two-dimensional method, so the description regarding the solutions
of the subsequent normal equations is omitted.
In step S5327, the approximation function generating unit 5326
determines regarding whether or not the processing of all the
pixels has been completed.
In step S5327, in the event that determination is made that the
processing of all the pixels has not been completed, the processing
returns to step S5322, wherein the subsequent processing is
repeatedly performed. That is to say, the pixels that have not
become a pixel of interest are sequentially taken as a pixel of
interest, and the processing in step S5322 through S5327 is
repeatedly performed.
In the event that the processing of all the pixels has been
completed (in step S5327, in the event that determination is made
that the processing of all the pixels has been completed), the
estimating processing of the actual world 1 ends.
As a result, for example, as shown in FIG. 380, the levels
(discontinuous functions) w.sub.1 through w.sub.5 serving as the
respective features are set for each rod-shaped region drawn with a
thick line in the direction of continuity (the velocity in the X
direction is v.sub.x, and the velocity in the Y direction is
v.sub.y), and the approximation function of the actual world is
estimated. In this case, with each rod-shaped region, the
cross-sectional size thereof as to the X-Y plane is d.times.e.
Also, the rod-shaped regions drawn with a fine line indicate the
case in which the velocity in the Y direction is v.sub.y=0. That is
to say, in the event of simply moving in the horizontal direction,
the rod-shaped regions to which the respective levels wi are set
keep the parallel relationship as to the X-t plane. This can be
applied to the case in which the velocity in the X direction is
v.sub.x=0. That is to say, in this case, each rod-shaped region
keeps the parallel relationship as to the Y-t plane.
Further, in the event that there is no change in the temporal
direction but continuity on the X-Y plane, the rod-shaped region
for each function keeps a position in parallel to the X-Y plane. In
other words, in the event that there is no change in the temporal
direction but continuity on the X-Y plane, there is a fine line or
two-valued edge.
Also, description has been made regarding the case in which each
region to which a function is discontinuously set is disposed in
the two-dimensional space (rod-shaped regions are disposed so as to
make up a plane) so far, but as shown in FIG. 381, each region may
be disposed within the three-dimensional space of XYT in a
stereoscopic manner, for example.
Also, with the above examples, description has been made regarding
the case in which the constant features w.sub.i are set as a
discontinuous function for each region, but the sameness may be
realized even in the event of employing a non-constant continuous
function. That is to say, for example, as shown in FIG. 382, when a
function as to the X direction is employed, an arrangement may be
made wherein the features w.sub.1 is set to w.sub.1=f.sub.0(x) with
the region of X0.ltoreq.X<X1 in the image, and the features
w.sub.2 is set to w.sub.2=f.sub.1(x) with the region of
X1.ltoreq.X<X2 in the image. Even an continuous function may be
set as a different function for each region. In this case, a
polynomial approximation function or a function other than that may
be employed as a function to be set.
Further, in the event that constant features w.sub.i are set for
each region as a discontinuous function, a function, which cannot
be continued at all at each region, may be set. That is to say, for
example, as shown in FIG. 383, when a function as to the X
direction is employed, an arrangement may be made wherein the
features w.sub.1 is set to w.sub.1=f.sub.0(x) with the region of
X0.ltoreq.X<X1 in the image, and the features w.sub.2 is set to
w.sub.2=f.sub.1(x) with the region of X1.ltoreq.X<X2 in the
image, whereby the same processing can be performed even if the
respective functions (e.g., f.sub.0(x) and f.sub.1(x)) are
discontinuous. In this case, a polynomial approximation function or
a function other than that may be employed as a function to be
set.
Thus, the actual world estimating unit 102 shown in FIG. 377 can
set an approximation function of the actual world by setting a
function discontinuously for each rod-shaped region in the
direction of continuity (angle or movement (the direction of
velocity which can be obtained from movement)) in the event of
setting each pixel value with a discontinuous function.
Next, description will be made regarding the image generating unit
103, which generates an image based on the actual world estimation
information estimated by the actual world estimating unit 102 shown
in the above FIG. 369.
The image generating unit 103 shown in FIG. 384 comprises an actual
world estimation information acquiring unit 5341, weighting
calculating unit 5342, and pixel generating unit 5343.
The actual world estimation information acquiring unit 5341
acquires features serving as the actual world estimation
information output from the actual world estimating unit 102 shown
in FIG. 369, i.e., a function (approximation function f(x) made up
of a discontinuous function), which sets a pixel value set for each
region divided in the direction of continuity, and outputs this to
the weighting calculating unit 5342.
The weighting calculating unit 5342 calculates the area ratio of
each region included in the pixel to be generated as weight based
on the information of the regions divided in the direction of
continuity, which is the actual world estimation information input
from the actual world estimation information acquiring unit 5341,
outputs the calculated results to the pixel generating unit 5343 as
well as the information of functions set for each region, which is
input from the actual world estimation information acquiring unit
5341.
The pixel generating unit 5343 obtains a level based on the
information of weight calculated based on the area ratio for each
region included in the pixel to be generated, which is input from
the weighting calculating unit, and the function (approximation
function f(x) made up of a discontinuous function) of the level set
for each region, obtains the sum of products of the level and
weight obtained for each pixel to be generated, and outputs this as
the pixel value of the pixel.
Next, description will be made regarding the image generating
processing by the image generating unit 103 shown in FIG. 384 with
reference to the flowchart shown in FIG. 385.
In step S5341, the actual world estimation information acquiring
unit 5341 acquires the actual world estimation information
(approximation function f(x) made up of a discontinuous function)
input from the actual world estimating unit 102 shown in FIG. 369,
and outputs this to the weighting calculating unit 5342.
The weighting calculating unit 5342 sets a pixel to be generated in
step S5342, obtains the area ratio as to the pixel to be generated
for each set region included in the pixel to be generated based on
the input actual world estimation information in step S5343,
calculates this as weight for each region, and outputs this to the
pixel generating unit 5343 as well as the function, which sets a
level for each region input from the actual world estimation
information acquiring unit 5341.
Description will be made regarding the case in which features are
set, as shown in FIG. 386, for example. Let us say that the pixels
of the input image are illustrated with fine-line grids, and a
pixel to be generated is illustrated with thick-line grids. That is
to say, in this case, a quadruple-density pixel is generated. Also,
let us say that five regions set in a band shape having a slope up
to right as to a pixel array illustrated with w.sub.1 through
w.sub.5 are regions set in the direction of continuity, and the
level of each region is w.sub.1 through w.sub.5.
In the event that the pixel painted in a hatched shape shown in
FIG. 386 is assumed to be a pixel of interest to be generated, the
pixel of interest extends over the regions w.sub.3 and w.sub.4, and
accordingly, when the areas occupied by each region within the
pixel of interest are m.sub.1 and m.sub.2 respectively, as for the
weight to be generated, when the area of a pixel to be generated is
m, the weight of the region w.sub.3 becomes m.sub.1/m, and the
weight of the region w.sub.4 becomes m.sub.2/m respectively. Thus,
the weighting calculating unit 5342 outputs the information of
weight obtained for each region, and the information of a function,
which sets the level of each region, to the pixel generating unit
5343.
In step S5344, the pixel generating unit 5343 determines a pixel
value based on the weight for each region which the pixel of
interest extends over, input from the weighting calculating unit
5342, and the level for each region, and generates a pixel.
That is to say, in the event of the pixel of interest described
with reference to FIG. 386, the pixel generating unit 5343 acquires
the information that the region w.sub.3 is m.sub.1/m, and the
region w.sub.4 is m.sub.2/m as each weight information. Further,
the pixel generating unit 5343 obtains the sum of products with the
level for each region acquired at the same time to determine a
pixel value, and generates a pixel.
That is to say, for example, in the event that the approximation
function, which determines the levels of the regions w.sub.3 and
w.sub.4, are w.sub.3 and w.sub.4 (both are constants), a pixel
value such as shown in the following Expression (297) is determined
by obtaining the sum of products with weight.
P=w.sub.3.times.m.sub.1/m+w.sub.4.times.m.sub.2/m (297)
In step S5345, the actual world estimation information acquiring
unit 5341 determines regarding whether or not the processing has
been completed as to all of the pixels of the image to be
generated, and in the event that determination is made that the
processing has not been completed as to all of the pixels, the
processing returns to step S5342, wherein the subsequent processing
is repeatedly performed. In other words, the processing in steps
S5342 through S5345 is repeatedly performed until determination is
made that the processing has been completed as to all of the
pixels.
In step S5345, in the event that determination is made that the
processing has been completed as to all of the pixels, the
processing thereof ends.
That is to say, for example, in the event that an object moves in
the horizontal direction, temporally in the right direction, it has
been known that as for actual change in a pixel value in the X-T
space in the actual world, regions indicating the same pixel value
level continue in the direction of continuity, as shown in A of
FIG. 387. Consequently, upon a higher density pixel being generated
using the model such as shown in B of FIG. 387, the shape of the
pixel cannot express actual linear movement having a slope up to
the right, and accordingly, for example, when attempting to
generate an enlarged image, an accurate pixel value cannot be
reflected upon the pixel generation of the enlarged image around a
boundary where a pixel value changes due to change in a pixel value
disposed geometrically in a staircase pattern.
Conversely, with the model in estimation of an approximation
function of the actual world by the actual world estimating unit
102 shown in FIG. 369, as shown in C of FIG. 387, a model, which is
faithful regarding actual movement, is generated in the direction
of continuity, and accordingly, change in a pixel level or less can
be accurately expressed, thereby enabling a high-density pixel used
for an enlarged image to be accurately generated, for example.
According to the above processing, a pixel can be generated taking
a light intensity distribution in a region of a pixel level or less
into consideration, and it becomes possible to generate a higher
density pixel, thereby enabling an enlarge image to be generated
photographically, for example.
Next, description will be made regarding the image generating unit
103, which generates an image based on the actual world estimation
information estimated by the actual world estimating unit 102 shown
in the above FIG. 377, with reference to FIG. 388.
The image generating unit 103 shown in FIG. 388 comprises an actual
world estimation information acquiring unit 5351, weighting
calculating unit 5352, and pixel generating unit 5353.
The actual world estimation information acquiring unit 5351
acquires features serving as the actual world estimation
information output from the actual world estimating unit 102 shown
in FIG. 377, i.e., a function (approximation function f(x) made up
of a discontinuous function), which sets a pixel value set for each
region divided in the direction of continuity, and outputs this to
the weighting calculating unit 5352.
The weighting calculating unit 5352 calculates the volume ratio of
each region included in the pixel to be generated as weight based
on the information of the regions divided in the direction of
continuity, which is the actual world estimation information input
from the actual world estimation information acquiring unit 5351,
and outputs the calculated results to the pixel generating unit
5353 as well as the information of functions set for each region,
which is input from the actual world estimation information
acquiring unit 5351.
The pixel generating unit 5353 obtains a level based on the
information of weight calculated based on the volume ratio for each
region included in the pixel to be generated, which is input from
the weighting calculating unit, and the function (approximation
function f(x) made up of a discontinuous function) of the level set
for each region, obtains the sum of products of the level and
weight obtained for each pixel to be generated, and outputs this as
the pixel value of the pixel thereof.
Next, description will be made regarding the image generating
processing by the image generating unit 103 shown in FIG. 388 with
reference to the flowchart shown in FIG. 389.
In step S5351, the actual world estimation information acquiring
unit 5351 acquires the actual world estimation information
(approximation function f(x) made up of a discontinuous function)
input from the actual world estimating unit 102 shown in FIG. 377,
and outputs this to the weighting calculating unit 5352.
The weighting calculating unit 5342 sets a pixel to be generated in
step S5352, obtains the volume ratio as to the pixel to be
generated for each set region included in the pixel to be generated
based on the input actual world estimation information in step
S5353, calculates this as weight for each region, and outputs this
to the pixel generating unit 5353 as well as the function, which
sets a level for each region input from the actual world estimation
information acquiring unit 5351.
For example, as shown in FIG. 390, let us say that a pixel of
interest is set as a pixel to be generated within the
three-dimensional space of the X direction, Y direction, and frame
direction T. Note that in FIG. 390, a cube expressed with a thick
line is the pixel of interest. Also, cubes drawn with a fine line
represent pixels adjacent to the pixel of interest.
Description will be made regarding the case in which features are
set, as shown in FIG. 391, for example. Let us say that three
regions set in a rod shape illustrated with w.sub.1 through w.sub.3
are regions set in the direction of continuity, and the level of
each region is w.sub.1 through w.sub.3.
As shown in FIG. 391, the pixel of interest extends over the
regions w.sub.3 through w.sub.4, and accordingly, when the volume
occupied by each region within the pixel of interest are M.sub.1
through M.sub.3 respectively, as for the weight to be generated,
when the volume of a pixel to be generated is M, the weight of the
region w.sub.1 becomes M.sub.1/M, the weight of the region w.sub.2
becomes M.sub.2/M, and the weight of the region w.sub.3 becomes
M.sub.3/M respectively. Thus, the weighting calculating unit 5342
outputs the information of weight obtained for each region, and the
information of a function, which sets the level of each region, to
the pixel generating unit 5353.
In step S5354, the pixel generating unit 5353 determines a pixel
value based on the weight for each region which the pixel of
interest extends over, input from the weighting calculating unit
5342, and the level for each region, and generates a pixel.
That is to say, in the event of the pixel of interest described
with reference to FIG. 391, the pixel generating unit 5353 acquires
the information that the region w.sub.1 is M.sub.1/M, the region
w.sub.2 is M.sub.2/M, and the region w.sub.3 is M.sub.3/M as each
weight information. Further, the pixel generating unit 5353 obtains
the sum of products with the level for each region acquired at the
same time to determine a pixel value, and generates a pixel.
That is to say, for example, in the event that the approximation
function, which determines the levels of the regions w.sub.1
through w.sub.3, are w.sub.1 through w.sub.3 (all is a constant), a
pixel value such as shown in the following Expression (298) is
determined by obtaining the sum of products with weight.
P=w.sub.1.times.M.sub.1/M+w.sub.2.times.M.sub.2/M+w.sub.3.times.M.sub.3/M
(298)
In step S5355, the actual world estimation information acquiring
unit 5351 determines regarding whether or not the processing has
been completed as to all of the pixels of the image to be
generated, and in the event that determination is made that the
processing has not been completed as to all of the pixels, the
processing returns to step S5352, wherein the subsequent processing
is repeatedly performed. In other words, the processing in steps
S5352 through S5355 is repeatedly performed until determination is
made that the processing has been completed as to all of the
pixels.
In step S5355, in the event that determination is made that the
processing has been completed as to all of the pixels, the
processing thereof ends.
A through D of FIG. 392 illustrate the processing results in the
case of generating a 16-powered density (quadruple density in the
horizontal direction and in the vertical direction respectively)
pixel as to the original image. A of FIG. 392 illustrates the
original image, B of FIG. 392 illustrates the processing result by
the conventional class classification adaptation processing, C of
FIG. 392 illustrates the processing result by the approximation
function of the actual world made up of the above polynomial, and
further, D of FIG. 392 illustrates the processing result by the
approximation function of the actual world made up of a
discontinuous function respectively.
With the processing result by the approximation function of the
actual world made up of a discontinuous function, it can be
understood that a clear image with little blurring similar to the
original image is generated.
Also, FIG. 393 illustrates, with the high density original image, a
comparison between the processing result by the approximation
function in the actual world made up of the above polynomial and
the approximation function in the actual world made up of a
discontinuous function after average pixel values of 4 pixels in
the horizontal direction.times.4 pixels in the vertical direction
are obtained, and further the space resolution is reduced to 1/16
with the pixel values of the 16 pixels thereof serving as the
obtained average pixel values. Note that in FIG. 393, a solid line
represents the original image, a dotted line represents the
processing result by the approximation function in the actual world
made up of a polynomial, a single-dot broken line represents the
processing result by the approximation function in the actual world
made up of a discontinuous function. Also, the horizontal axis in
the drawing represents coordinate positions in the X direction, and
the vertical axis represents pixel values.
It can be understood that the processing result by the
approximation function in the actual world made up of a
discontinuous function is more identical to the original image at
x=651 through 655, and reproduces a pixel value accurately in
generation of a 16-powered density pixel as compared to the
processing result by the approximation function in the actual world
made up of a polynomial.
According to the above processing, a pixel can be generated taking
a light intensity distribution in a region of the pixel level or
less into consideration, and a higher density pixel can be
accurately generated, thereby enabling an enlarged image to be
generated clearly, for example.
Further, as described above, according to the method for setting an
approximation function in the actual world made up of a
discontinuous function, even if movement blurring occurs in an
image, this can be removed.
Now, description will be made regarding an input image and movement
blurring with reference to FIG. 394 through FIG. 409.
FIG. 394 is a diagram for describing imaging by the sensor 2. The
sensor 2 comprises, for example, a CCD video camera including a CCD
(Charge-Coupled Device) area sensor serving as a solid-state
imaging device, and the like. An object corresponding to the
foreground in the real world moves between an object corresponding
to the background in the real world and the sensor, e.g.,
horizontally from the left side to the right side in the
drawing.
The sensor 2 takes an image of an object corresponding to the
foreground as well as an object corresponding to the background.
The sensor 2 outputs the taken image in increments of one frame.
For example, the sensor 2 outputs an image of 30 frames per second.
In this case, the exposure time of the sensor 2 can be made to be
1/30 seconds. The exposure time is the time from the sensor 2
starting conversion of input light into electric charge, to ending
of the conversion of input light into electric charge. Hereafter,
the exposure time will also be called shutter time.
FIG. 395 is a diagram describing placement of a pixel. In FIG. 395,
A through I denote individual pixels. The pixels are placed on a
plane corresponding to an image. A single detecting element
corresponding to a single pixel is placed on the sensor 2. At the
time of the sensor 2 taking an image, the one detecting element
outputs one pixel value corresponding to the one pixel making up
the image. For example, the position in the X direction X of the
detecting element corresponds to the horizontal position on the
image, and the position in the Y direction of the detecting element
corresponds to the vertical position on the image.
As shown in FIG. 396, the detecting device which is a CCD for
example, converts input light into electric charge during a period
corresponding to the shutter time, and accumulates the converted
charge. The amount of charge is approximately proportionate to the
intensity of input light, and the amount of time that light is
input. That is to say, the detecting device integrates the light to
be input, and accumulates a change of an amount corresponding to
the integrated light during a period corresponding to the shutter
time.
The charge accumulated in the detecting device is converted into a
voltage value by an unshown circuit, the voltage value is further
converted into a pixel value such as digital data or the like, and
is output. Accordingly, the individual pixel values output from the
sensor 2 have a value projected on one-dimensional space, which is
the result of integrating the portion having time-space expanse of
an object corresponding to the foreground or background with regard
to the time direction of the shutter time.
FIG. 397 is a diagram for describing an image obtained by taking an
object corresponding to the moving foreground and an object
corresponding to the background. A in FIG. 397 illustrates an image
obtained by taking an object accompanying movement, and an object
corresponding to the still background. With the example shown in A
in FIG. 397, the object corresponding to the foreground moves
horizontally from the left to the right as to the screen.
B in FIG. 397 is a model diagram wherein a pixel value
corresponding to a single line of the image shown in A in FIG. 397
is extended in the time direction. The horizontal direction of B in
FIG. 397 corresponds to the spatial direction X of A in FIG.
397.
With the pixels in the background region, the pixel values thereof
comprise the background components alone, i.e., only the components
of the image corresponding to the background object. With the
pixels in the foreground region, the pixel values thereof comprise
the foreground components alone, i.e., only the components of the
image corresponding to the foreground object.
With the pixels in the mixed region, the pixel values thereof
comprise the foreground components and the background components.
The mixed region can also be referred to as a strain region since
the pixel values thereof comprise the foreground components and the
background components. The mixed region is further classified into
a covered background region and an uncovered background region.
The covered background region is a mixed region in a position
corresponding to the front end portion in the direction of movement
of the foreground object as to the foreground region, i.e., a
region of which the background components are covered up by the
foreground according to elapsed time.
On the other hand, the uncovered background region is a mixed
region in a position corresponding to the rear end portion in the
direction of movement of the foreground object as to the foreground
region, i.e., a region of which the background components emerge
according to elapsed time.
FIG. 398 is a diagram for describing the background region,
foreground region, mixed region, covered background region, and
uncovered background region, as described above. In the event of
correlating those with the image shown in FIG. 397, the background
region is a still portion, the foreground region is a movement
portion, the covered background region of the mixed region is a
portion, which is changed from the background to the foreground,
and the uncovered background region of the mixed region is a
portion, which is changed from the foreground to the
background.
FIG. 399 is a model diagram wherein the pixel values of the pixels
arrayed adjacently in a row in the image obtained by taking an
object corresponding to the still foreground, and an object
corresponding to the still background. For example, pixels arrayed
on one line of the screen can be selected as pixels arrayed
adjacently in a row.
The pixel values F01 through F04 shown in FIG. 399 are the pixel
values of the pixels corresponding to the still foreground object.
The pixel values B01 through B04 shown in FIG. 399 are the pixel
values of the pixels corresponding to the still background
object.
The vertical direction in FIG. 399 corresponds to time, wherein
time elapses from top down in the drawing. The upper side position
of a rectangle in FIG. 399 corresponds to a point-in-time for the
sensor 2 starting conversion of the input light to electric charge,
and the lower side position of a rectangle in FIG. 399 corresponds
to a point-in-time for the sensor 2 completing conversion of the
input light to electric charge. That is to say, the distance from
the upper side to the lower side of a rectangle corresponds to
shutter time.
Description will be made below regarding the case in which shutter
time and a frame interval are the same as an example.
The horizontal direction in FIG. 399 corresponds to the spatial
direction X described in FIG. 397. More specifically, with the
example shown in FIG. 399, the distance from the left side of the
rectangle denoted with "F01" to the right side of the rectangle
denoted with "B04" in FIG. 399 is octuple a pixel pitch, i.e.,
corresponds to the interval of consecutive eight pixels.
In the event that the foreground object and background object are
still, the light to be input to the sensor 2 does not change during
a period corresponding to the shutter time.
Now, the period corresponding to the shutter time is divided into
two or more periods having the same length. The number of virtual
division is set corresponding to amount-of-movement v within the
shutter time of an object corresponding to the foreground. For
example, as shown in FIG. 400, the number of virtual division is
set to four corresponding to the amount-of-movement v, which is
four, so that the period corresponding to the shutter time is
divided into four.
The top line in FIG. 400 corresponds to the first period following
the shutter opening. The second line from the top in the drawing
corresponds to the second period following the shutter opening. The
third line from the top in the drawing corresponds to the third
period following the shutter opening. The fourth line from the top
in the drawing corresponds to the fourth period following the
shutter opening.
Hereafter, the shutter time divided corresponding to the
amount-of-movement v is also referred to as shutter time/v.
When the object corresponding to the foreground is still, the light
to be input to the sensor 2 does not change, so that the foreground
component F01/v is equal to a value obtained by dividing the pixel
value F01 by the number of virtual division. Similarly, when the
object corresponding to the foreground is still, the foreground
component F02/v is equal to a value obtained by dividing the pixel
value F02 by the number of virtual division, the foreground
component F03/v is equal to a value obtained by dividing the pixel
value F03 by the number of virtual division, and the foreground
component F04/v is equal to a value obtained by dividing the pixel
value F04 by the number of virtual division.
When the object corresponding to the background is still, the light
to be input to the sensor 2 does not change, so that the background
component B01/v is equal to a value obtained by dividing the pixel
value B01 by the number of virtual division. Similarly, when the
object corresponding to the background is still, the background
component B02/v is equal to a value obtained by dividing the pixel
value B02 by the number of virtual division, the background
component B03/v is equal to a value obtained by dividing the pixel
value B03 by the number of virtual division, and the background
component B04/v is equal to a value obtained by dividing the pixel
value B04 by the number of virtual division.
That is to say, in the event that the object corresponding to the
foreground is still, the light corresponding to the foreground
object, which is input to the sensor 2, does not change during the
period corresponding to the shutter time, so that the foreground
component F01/v corresponding to the first shutter time/v following
the shutter opening, the foreground component F01/v corresponding
to the second shutter time/v following the shutter opening, the
foreground component F01/v corresponding to the third shutter
time/v following the shutter opening, and the foreground component
F01/v corresponding to the fourth shutter time/v following the
shutter opening, become the same value. The F02/v through F04/v
have the same relationship as the F01/v.
In the event that the object corresponding to the background is
still, the light corresponding to the background object, which is
input to the sensor 2, does not change during the period
corresponding to the shutter time, so that the background component
B01/v corresponding to the first shutter time/v following the
shutter opening, the background component B01/v corresponding to
the second shutter time/v following the shutter opening, the
background component B01/v corresponding to the third shutter
time/v following the shutter opening, and the background component
B01/v corresponding to the fourth shutter time/v following the
shutter opening, become the same value. The B02/v through B04/v
have the same relationship as the B01/v.
Next, description will be made regarding the case in which an
object corresponding the foreground moves, and an object
corresponding to the background is still.
FIG. 401 is a model diagram wherein the pixel values of the pixels
on one line including a covered background region are extended in
the time direction, in the event that an object corresponding to
the foreground moves to the right side in the drawing. In FIG. 401,
the foreground amount-of-movement v is four. We can assume that one
frame is a short period, so the object corresponding to the
foreground is a stiffness member, and moves at constant velocity.
In FIG. 401, the image of an object corresponding to the foreground
moves so as to be displayed with a shift of four pixels on the
right side in the next frame on the basis of a certain frame.
In FIG. 401, the pixel on the leftmost side through the fourth
pixel from the left belong to a foreground region. In FIG. 401, the
fifth pixel from the left through the seventh pixel from the left
belong to a mixed region serving as a covered background region. In
FIG. 401, the pixel on the rightmost side belongs to a background
region.
The object corresponding to the foreground moves as time elapses so
as to cover up the object corresponding to the background, so that
the components included in the pixel values of the pixels belonged
to the covered background region are switched to the foreground
components from the background components at a certain point of
time during the period corresponding to the shutter time.
For example, a pixel value M appended with a thick-line frame in
FIG. 401 is represented with Expression (299).
M=B02/v+B02/v+F07/v+F06/v (299)
For example, the fifth pixel from the left includes background
components corresponding to one shutter time/v, and includes
foreground components corresponding to three sets of shutter
time/v, and accordingly, the mixed ratio .alpha. of the fifth pixel
from the left is 1/4. The sixth pixel from the left includes
background components corresponding to two sets of shutter time/v,
and includes foreground components corresponding to two sets of
shutter time/v, and accordingly, the mixed ratio .alpha. of the
sixth pixel from the left is 1/2. The seventh pixel from the left
includes background components corresponding to three sets of
shutter time/v, and includes foreground components corresponding to
one shutter time/v, and accordingly, the mixed ratio .alpha. of the
seventh pixel from the left is 3/4.
We can assume that the object corresponding to the foreground is a
stiffness member, and the foreground image moves at constant
velocity so as to be displayed with a shift of four pixels on the
right side in the next frame, so that, for example, the foreground
component F07/v of the first shutter time/v following the shutter
opening of the fourth pixel from the left in FIG. 401 is equal to
the foreground component corresponding to the second shutter time/v
following the shutter opening of the fifth pixel from the left in
FIG. 401. Similarly, the foreground component F07/v is equal to the
foreground component corresponding to the third shutter time/v
following the shutter opening of the sixth pixel from the left in
FIG. 401, and the foreground component corresponding to the fourth
shutter time/v following the shutter opening of the seventh pixel
from the left in FIG. 401 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and the foreground image moves at constant
velocity so as to be displayed with a shift of four pixels on the
right side in the next frame, so that, for example, the foreground
component F06/v of the first shutter time/v following the shutter
opening of the third pixel from the left in FIG. 401 is equal to
the foreground component corresponding to the second shutter time/v
following the shutter opening of the fourth pixel from the left in
FIG. 401. Similarly, the foreground component F06/v is equal to the
foreground component corresponding to the third shutter time/v
following the shutter opening of the fifth pixel from the left in
FIG. 401, and the foreground component corresponding to the fourth
shutter time/v following the shutter opening of the sixth pixel
from the left in FIG. 401 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and the foreground image moves at constant
velocity so as to be displayed with a shift of four pixels on the
right side in the next frame, so that, for example, the foreground
component F05/v of the first shutter time/v following the shutter
opening of the second pixel from the left in FIG. 401 is equal to
the foreground component corresponding to the second shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 401. Similarly, the foreground component F05/v is equal to the
foreground component corresponding to the third shutter time/v
following the shutter opening of the fourth pixel from the left in
FIG. 401, and the foreground component corresponding to the fourth
shutter time/v following the shutter opening of the fifth pixel
from the left in FIG. 401 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and the foreground image moves at constant
velocity so as to be displayed with a shift of four pixels on the
right side in the next frame, so that, for example, the foreground
component F04/v of the first shutter time/v following the shutter
opening of the pixel on the leftmost side in FIG. 401 is equal to
the foreground component corresponding to the second shutter time/v
following the shutter opening of the second pixel from the left in
FIG. 401. Similarly, the foreground component F04/v is equal to the
foreground component corresponding to the third shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 401, and the foreground component corresponding to the fourth
shutter time/v following the shutter opening of the fourth pixel
from the left in FIG. 401 respectively.
The state of the foreground region corresponding to such a moving
object is movement blurring. Also, the foreground region
corresponding to a moving object thus includes movement blurring,
so can be referred to as a strain region.
FIG. 402 is a model diagram wherein the pixel values of the pixels
on one line including an uncovered background region are extended
in the time direction, in the event that the foreground moves to
the right side in the drawing. In FIG. 402, the foreground
amount-of-movement v is four. We can assume that one frame is a
short period, so the object corresponding to the foreground is a
stiffness member, and moves at constant velocity. In FIG. 402, the
image of an object corresponding to the foreground moves with a
shift of four pixels on the right side in the next frame on the
basis of a certain frame.
In FIG. 402, the pixel on the leftmost side through the fourth
pixel from the left belong to a background region. In FIG. 402, the
fifth pixel from the left through the seventh pixel from the left
belong to a mixed region serving as an uncovered background region.
In FIG. 402, the pixel on the rightmost side belongs to a
foreground region.
The object corresponding to the foreground, which has covered up
the object corresponding to the background, moves as time elapses
so as to be removed from front of the object corresponding to the
background, so that the components included in the pixel values of
the pixels belonged to the uncovered background region are switched
to the background components from the foreground components at a
certain point of time during the period corresponding to the
shutter time.
For example, a pixel value M' appended with a thick-line frame in
FIG. 402 is represented with Expression (300).
M'=F02/v+F01/v+B26/v+B26/v (300)
For example, the fifth pixel from the left includes background
components corresponding to three sets of shutter time/v, and
includes foreground components corresponding to one shutter time/v,
and accordingly, the mixed ratio .alpha. of the fifth pixel from
the left is 3/4. The sixth pixel from the left includes background
components corresponding to two sets of shutter time/v, and
includes foreground components corresponding to two sets of shutter
time/v, and accordingly, the mixed ratio .alpha. of the sixth pixel
from the left is 1/2. The seventh pixel from the left includes
background components corresponding to one shutter time/v, and
includes foreground components corresponding to three sets of
shutter time/v, and accordingly, the mixed ratio .alpha. of the
seventh pixel from the left is 1/4.
If Expression (299) and Expression (300) are more generalized, the
pixel value M is represented with Expression (301).
.alpha..times..times. ##EQU00177##
Here, .alpha. represents the mixed ratio. B represents a background
pixel value, and Fi/v represents a foreground component.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and also the
amount-of-movement v is four, and accordingly, for example, the
foreground component F01/v of the first shutter time/v following
the shutter opening of the fifth pixel from the left in FIG. 402 is
equal to the foreground component corresponding to the second
shutter time/v following the shutter opening of the sixth pixel
from the left in FIG. 402. Similarly, the F01/v is equal to the
foreground component corresponding to the third shutter time/v
following the shutter opening of the seventh pixel from the left in
FIG. 402, and the foreground component corresponding to the fourth
shutter time/v following the shutter opening of the eighth pixel
from the left in FIG. 402 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and also the
number of virtual division is four, so that, for example, the
foreground component F02/v of the first shutter time/v following
the shutter opening of the sixth pixel from the left in FIG. 402 is
equal to the foreground component corresponding to the second
shutter time/v following the shutter opening of the seventh pixel
from the left in FIG. 402. Similarly, the foreground component
F02/v is equal to the foreground component corresponding to the
third shutter time/v following the shutter opening of the eighth
pixel from the left in FIG. 402.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and also the
amount-of-movement v is four, so that, for example, the foreground
component F03/v of the first shutter time/v following the shutter
opening of the seventh pixel from the left in FIG. 402 is equal to
the foreground component corresponding to the second shutter time/v
following the shutter opening of the eighth pixel from the left in
FIG. 402.
With description in FIG. 400 through FIG. 402, description has been
made on condition that the number of virtual division is four, but
the number of virtual division corresponds to an amount-of-movement
v. The amount-of-movement v generally corresponds to the movement
speed of an object corresponding to the foreground. For example,
when an object corresponding to the foreground is moving so as to
be displayed with a shift of four pixels on the right side in the
next frame on the basis of a certain frame, the amount-of-movement
v is set to four. The number of virtual division corresponds to the
amount-of-movement v, and is set to four. Similarly, for example,
when an object corresponding to the foreground is moving so as to
be displayed with a shift of six pixels on the left side in the
next frame on the basis of a certain frame, the amount-of-movement
v is set to six. The number of virtual is set to six.
FIG. 403 and FIG. 404 illustrate the relationship between the above
foreground region, background region, and mixed region made up of a
covered background region or uncovered background region, and the
foreground components and background components corresponding to
the divided shutter time.
FIG. 403 illustrates an example wherein the pixels in the
foreground region, background region, and mixed region are
extracted from an image including the foreground corresponding to
an object moving in front of the still background. With the example
shown in FIG. 403, an object corresponding to the foreground is
moving horizontally as to a screen.
A frame.sup.#n+1 is the subsequent frame of a frame.sup.#n, and a
frame.sup.#n+2 is the subsequent frame of the frame.sup.#n+1.
FIG. 404 illustrates a model wherein the pixels in the foreground
region, background region, and mixed region, which are extracted
from any one of the frame.sup.#n through frame.sup.#n+2, are
extracted, the amount-of-movement is set to four, and the pixel
values of the extracted pixels are extended in the time
direction.
The pixel values in the foreground region comprise four different
foreground components corresponding to the period of the shutter
time/v since the object corresponding to the foreground moves. For
example, the pixel positioned on the leftmost side of the pixels in
the foreground region shown in FIG. 404 comprise F01/v, F02/v,
F03/v, and F04/v. That is to say, the pixels in the foreground
region include movement blurring.
The object corresponding to the background is still, so the light
corresponding to the background, which is input to the sensor 2,
does not change in the period corresponding to the shutter time. In
this case, the pixel values in the background region do not include
movement blurring.
The pixel values of the pixels belonged to the mixed region made up
of the covered background region or uncovered background region
comprise foreground components and background components.
Next, description will be made regarding a model wherein when an
image corresponding to an object is moving, the pixel values of the
pixels, which are pixels adjacently arrayed in one row in multiple
frames, on the same position on the frames are extended in the time
direction. For example, when an image corresponding to an object is
moving horizontally as to a screen, the pixels arrayed on one line
of a screen can be selected as pixels adjacently arrayed in one
row.
FIG. 405 is a model diagram wherein the pixel values of the pixels,
which are pixels adjacently arrayed in one row of three frames of
an image obtained by taking an object corresponding to the still
background, on the same positions on the frames are extended in the
time direction. A frame.sup.#n is the subsequent frame of a
frame.sup.#n-1, and a frame.sup.#n+1 is the subsequent frame of the
frame.sup.#n. The other frames are referred in the same way.
The pixel values B01 through B12 shown in FIG. 405 are the pixel
values of the pixels corresponding to the object of the still
background. The object corresponding to the background is still, so
the pixel values of the corresponding pixels do not change in the
frame.sup.#n-1 through frame.sup.#n+1. For example, the pixel in
the frame.sup.#n and the pixel in the frame.sup.#+1, which
correspond to the position of the pixel having a pixel value B05 in
the frame.sup.#n-1, each have the pixel value B05.
FIG. 406 is a model diagram wherein the pixel values of the pixels,
which are pixels adjacently arrayed in one row of three frames of
an image obtained by taking an object corresponding to the
foreground moving to the right side in the drawing as well as an
object corresponding to the still background, on the same positions
on the frames are extended in the time direction. The model shown
in FIG. 406 includes a covered background region.
In FIG. 406, we can assume that the object corresponding to the
foreground is a stiffness member, and moves at constant velocity,
and the image of the foreground moves so as to be displayed with a
shift of four pixels on the right side in the next frame, and
accordingly, the amount-of-movement v of the foreground is four,
and the number of virtual division is four.
For example, the foreground component of the first shutter time/v
following the shutter opening of the pixel on the leftmost side on
the frame.sup.#n-1 in FIG. 406 becomes F12/v, and the foreground
component of the second shutter time/v following the shutter
opening of the second pixel from the left in FIG. 406 becomes F12/v
as well. The foreground component of the third shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 406, and the foreground component of the fourth shutter time/v
following the shutter opening of the fourth pixel from the left in
FIG. 406 become F12/v.
The foreground component of the second shutter time/v following the
shutter opening of the pixel on the leftmost side on the
frame.sup.#n-1 in FIG. 406 becomes F11/v, and the foreground
component of the third shutter time/v following the shutter opening
of the second pixel from the left in FIG. 406 becomes F11/v as
well. The foreground component of the fourth shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 406 becomes F11/v as well.
The foreground component of the third shutter time/v following the
shutter opening of the pixel on the leftmost side on the
frame.sup.#n-1 in FIG. 406 becomes F10/v, and the foreground
component of the fourth shutter time/v following the shutter
opening of the second pixel from the left in FIG. 406 becomes F10/v
as well. The foreground component of the fourth shutter time/v
following the shutter opening of the pixel on the leftmost side on
the frame.sup.#n-1 in FIG. 406 becomes F09/v.
The object corresponding the background is still, so the background
component of the first shutter time/v following the shutter opening
of the second pixel from the left on the frame.sup.#n-1 in FIG. 406
becomes B01/v. The background components of the first through third
shutter time/v following the shutter opening of the fourth pixel
from the left on the frame.sup.#n-1 in FIG. 406 become B03/v.
With the frame.sup.#n-1 in FIG. 406, the pixels on the leftmost
side belong to the foreground region, and the second through fourth
pixels from the left belong to the mixed region serving as a
covered background region.
The fifth through twelfth pixels from the left on the
frame.sup.#n-1 in FIG. 406 belong to the background region, and the
pixel values thereof become B04 through B11 respectively.
The first through fifth pixels from the left on the frame.sup.#n in
FIG. 406 belong to the foreground region. The foreground components
of the shutter time/v in the foreground region on the frame.sup.#n
are any one of F05/v through F12/v.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and the image of
the foreground moves so as to be displayed with a shift of four
pixels on the right side in the next frame, and accordingly, the
foreground component of the first shutter time/v following the
shutter opening of the fifth pixel from the left on the
frame.sup.#n in FIG. 406 becomes F12/v, the foreground component of
the second shutter time/v following the shutter opening of the
sixth pixel from the left in FIG. 406 becomes F12/v as well. The
foreground component of the third shutter time/v following the
shutter opening of the seventh pixel from the left in FIG. 406, and
the foreground component of the fourth shutter time/v following the
shutter opening of the eighth pixel from the left in FIG. 406
become F12/v.
The foreground component of the second shutter time/v following the
shutter opening of the fifth pixel from the left on the
frame.sup.#n in FIG. 406 becomes F11/v, and the foreground
component of the third shutter time/v following the shutter opening
of the sixth pixel from the left in FIG. 406 becomes F11/v as well.
The foreground component of the fourth shutter time/v following the
shutter opening of the seventh pixel from the left in FIG. 406
becomes F11/v.
The foreground component of the third shutter time/v following the
shutter opening of the fifth pixel from the left on the
frame.sup.#n in FIG. 406 becomes F10/v, and the foreground
component of the fourth shutter time/v following the shutter
opening of the sixth pixel from the left in FIG. 406 becomes F10/v
as well. The foreground component of the fourth shutter time/v
following the shutter opening of the fifth pixel from the left on
the frame.sup.#n in FIG. 406 becomes F09/v.
The object corresponding the background is still, so the background
component of the first shutter time/v following the shutter opening
of the sixth pixel from the left on the frame.sup.#n in FIG. 406
becomes B05/v. The background components of the first through
second shutter time/v following the shutter opening of the seventh
pixel from the left on the frame.sup.#n in FIG. 406 become B06/v.
The background components of the first through third shutter time/v
following the shutter opening of the eighth pixel from the left on
the frame.sup.#n in FIG. 406 become B07/v.
With the frame.sup.#n in FIG. 406, the sixth through eighth pixels
from the left belong to the mixed region serving as a covered
background region.
The ninth through twelfth pixels from the left on the frame.sup.#n
in FIG. 406 belong to the background region, and the pixel values
thereof become B08 through B11 respectively.
The ninth through twelfth pixels from the left on the
frame.sup.#n+1 in FIG. 406 belong to the foreground region. With
the foreground region of the frame.sup.#n+1, the foreground
components are any one of F01/v through F12/v.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and the image of
the foreground moves so as to be displayed with a shift of four
pixels on the right side in the next frame, and accordingly, the
foreground component of the first shutter time/v following the
shutter opening of the ninth pixel from the left on the
frame.sup.#n+1 in FIG. 406 becomes F12/v, the foreground component
of the second shutter time/v following the shutter opening of the
tenth pixel from the left in FIG. 406 becomes F12/v as well. The
foreground component of the third shutter time/v following the
shutter opening of the eleventh pixel from the left in FIG. 406,
and the foreground component of the fourth shutter time/v following
the shutter opening of the twelfth pixel from the left in FIG. 406
become F12/v.
The foreground component of the second shutter time/v following the
shutter opening of the ninth pixel from the left on the
frame.sup.#n+1 in FIG. 406 becomes F11/v, and the foreground
component of the third shutter time/v following the shutter opening
of the tenth pixel from the left in FIG. 406 becomes F11/v as well.
The foreground component of the fourth shutter time/v following the
shutter opening of the eleventh pixel from the left in FIG. 406
becomes F11/v.
The foreground component of the third shutter time/v following the
shutter opening of the ninth pixel from the left on the
frame.sup.#n+1 in FIG. 406 becomes F10/v, and the foreground
component of the fourth shutter time/v following the shutter
opening of the tenth pixel from the left in FIG. 406 becomes F10/v
as well. The foreground component of the fourth shutter time/v
following the shutter opening of the ninth pixel from the left on
the frame.sup.#n+1 in FIG. 406 becomes F09/v.
The object corresponding the background is still, so the background
component of the first shutter time/v following the shutter opening
of the tenth pixel from the left on the frame.sup.#n+1 in FIG. 406
becomes B09/v. The background components of the first through
second shutter time/v following the shutter opening of the eleventh
pixel from the left on the frame.sup.#n+1 in FIG. 406 become B10/v.
The background components of the first through third shutter time/v
following the shutter opening of the twelfth pixel from the left on
the frame.sup.#n+1 in FIG. 406 become B11/v.
With the frame.sup.#n+1 in FIG. 406, the tenth through twelfth
pixels from the left correspond to the mixed region serving as a
covered background region.
FIG. 407 is a model diagram of an image wherein the foreground
components are extracted from the pixel values shown in FIG.
406.
FIG. 408 is a model diagram wherein the pixel values of the pixels,
which are pixels adjacently arrayed in one row of three frames of
an image obtained by taking the foreground corresponding to an
object moving to the right side in the drawing as well as the still
background, on the same positions on the frames are extended in the
time direction. In FIG. 408 an uncovered background region is
included.
In FIG. 408, we can assume that the object corresponding to the
foreground is a stiffness member, and moves at constant velocity.
The object corresponding to the foreground moves so as to be
displayed with a shift of four pixels on the right side in the next
frame, and accordingly, the amount-of-movement v is four.
For example, the foreground component of the first shutter time/v
following the shutter opening of the pixel on the leftmost side on
the frame.sup.#n-1 in FIG. 408 becomes F13/v, and the foreground
component of the second shutter time/v following the shutter
opening of the second pixel from the left in FIG. 408 becomes F13/v
as well. The foreground component of the third shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 408, and the foreground component of the fourth shutter time/v
following the shutter opening of the fourth pixel from the left in
FIG. 408 become F13/v.
The foreground component of the first shutter time/v following the
shutter opening of the second pixel from the left on the
frame.sup.#n-1 in FIG. 408 becomes F14/v, and the foreground
component of the second shutter time/v following the shutter
opening of the third pixel from the left in FIG. 408 becomes F14/v
as well. The foreground component of the first shutter time/v
following the shutter opening of the third pixel from the left in
FIG. 408 becomes F15/v.
The object corresponding the background is still, so the background
components of the second through fourth shutter time/v following
the shutter opening of the pixel on the leftmost side on the
frame.sup.#n-1 in FIG. 408 become B25/v. The background components
of the third through fourth shutter time/v following the shutter
opening of the second pixel from the left on the frame.sup.#n-1 in
FIG. 408 become B26/v. The background components of the fourth
shutter time/v following the shutter opening of the third pixel
from the left on the frame.sup.#n-1 in FIG. 408 becomes B27/v.
With the frame.sup.#n-1 in FIG. 408, the leftmost pixel through the
third pixel belong to the mixed region serving as an uncovered
background region.
The fourth through twelfth pixels from the left on the
frame.sup.#n-1 in FIG. 408 belong to the foreground region. The
foreground components of the frame are any one of F13/v through
F24/v.
The leftmost pixel through fourth pixel from the left on the
frame.sup.#n in FIG. 408 belong to the background region, and the
pixel values thereof are B25 through B28 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and the image of
the foreground moves so as to be displayed with a shift of four
pixels on the right side in the next frame, and accordingly, the
foreground component of the first shutter time/v following the
shutter opening of the fifth pixel from the left on the
frame.sup.#n in FIG. 408 becomes F13/v, the foreground component of
the second shutter time/v following the shutter opening of the
sixth pixel from the left in FIG. 408 becomes F13/v as well. The
foreground component of the third shutter time/v following the
shutter opening of the seventh pixel from the left in FIG. 408, and
the foreground component of the fourth shutter time/v following the
shutter opening of the eighth pixel from the left in FIG. 408
become F13/v.
The foreground component of the first shutter time/v following the
shutter opening of the sixth pixel from the left on the
frame.sup.#n in FIG. 408 becomes F14/v, and the foreground
component of the second shutter time/v following the shutter
opening of the seventh pixel from the left in FIG. 408 becomes
F14/v as well. The foreground component of the first shutter time/v
following the shutter opening of the eighth pixel from the left in
FIG. 408 becomes F15/v.
The object corresponding the background is still, so the background
components of the second through fourth shutter time/v following
the shutter opening of the fifth pixel from the left on the
frame.sup.#n in FIG. 408 become B29/v. The background components of
the third through fourth shutter time/v following the shutter
opening of the sixth pixel from the left on the frame.sup.#n in
FIG. 408 become B30/v. The background components of the fourth
shutter time/v following the shutter opening of the seventh pixel
from the left on the frame.sup.#n in FIG. 408 becomes B31/v.
With the frame.sup.#n in FIG. 408, the fifth through seventh pixels
from the left belong to the mixed region serving as an uncovered
background region.
The eighth through twelfth pixels from the left on the frame.sup.#n
in FIG. 408 belong to the foreground region. The values
corresponding to the period of the shutter time/v in the foreground
region on the frame.sup.#n are any one of F13/v through F20/v.
The leftmost pixel through eighth pixel from the left on the
frame.sup.#n+1 in FIG. 408 belong to the background region, and the
pixel values thereof are B25 through B32 respectively.
We can assume that the object corresponding to the foreground is a
stiffness member, and moves at constant velocity, and the image of
the foreground moves so as to be displayed with a shift of four
pixels on the right side in the next frame, and accordingly, the
foreground component of the first shutter time/v following the
shutter opening of the ninth pixel from the left on the
frame.sup.#n+1 in FIG. 408 becomes F13/v, and the foreground
component of the second shutter time/v following the shutter
opening of the tenth pixel from the left in FIG. 408 becomes F13/v
as well. The foreground component of the third shutter time/v
following the shutter opening of the eleventh pixel from the left
in FIG. 408, and the foreground component of the fourth shutter
time/v following the shutter opening of the twelfth pixel from the
left in FIG. 408 become F13/v.
The foreground component of the first shutter time/v following the
shutter opening of the tenth pixel from the left on the
frame.sup.#n+1 in FIG. 408 becomes F14/v, and the foreground
component of the second shutter time/v following the shutter
opening of the eleventh pixel from the left in FIG. 408 becomes
F14/v as well. The foreground component of the first shutter time/v
following the shutter opening of the twelfth pixel from the left in
FIG. 408 becomes F15/v.
The object corresponding the background is still, so the background
components of the second through fourth shutter time/v following
the shutter opening of the ninth pixel from the left on the
frame.sup.#n+1 in FIG. 408 become B33/v. The background components
of the third through fourth shutter time/v following the shutter
opening of the tenth pixel from the left on the frame.sup.#n+1 in
FIG. 408 become B34/v. The background components of the fourth
shutter time/v following the shutter opening of the eleventh pixel
from the left on the frame.sup.#n+1 in FIG. 408 becomes B35/v.
With the frame.sup.#n+1 in FIG. 408, the ninth through eleventh
pixels from the left belong to the mixed region serving as an
uncovered background region.
The twelfth pixel from the left on the frame.sup.#n+1 in FIG. 408
belongs to the foreground region. The foreground components of the
shutter time/v in the foreground region on the frame.sup.#n+1 are
any one of F13/v through F16/v.
FIG. 409 is a model diagram of an image wherein the foreground
components are extracted from the pixel values shown in FIG.
408.
Description has been made so far regarding the input image and
movement blurring so far, and change in the components within a
pixel has been described with the number of virtual division, but
each component has the same configuration as the band-shaped
regions shown with the levels w.sub.1 through w.sub.5 positioned in
the right portion of FIG. 373 by setting the number of virtual
division to infinite, for example.
That is to say, it can be said that to set the levels as a
discontinuous function on the X-T plane (the same even on the X-Y
plane) for each region in the direction of continuity is to set
change in the components within the shutter time as a linear region
instead of the number of virtual division.
On this account, the mechanism for generating the above movement
blurring can be estimated by estimating the actual world using an
approximation function made up of a discontinuous function for each
region in the direction of continuity.
Accordingly, movement blurring may be essentially removed by
utilizing this property, i.e., by generating a pixel within one
shutter time (one pixel or less in the frame direction).
FIG. 410 is the comparison between the processing result in the
case of removing movement blurring by the class classification
adaptation processing, and the processing result in the case of
removing movement blurring using an approximation function in the
actual world obtained by setting a discontinuous function for each
region in the direction of continuity. Note that in FIG. 410, a
dotted line illustrates change in the pixel value in an input image
(image wherein movement blurring exists), a solid line illustrates
the processing result in the case of removing movement blurring by
the class classification adaptation processing, and a single-dot
broken line illustrates the processing result in the case of
removing movement blurring using an approximation function in the
actual world obtained by setting a discontinuous function for each
region in the direction of continuity. Further, the horizontal axis
represents coordinates in the X direction of the input image, and
the vertical axis represents pixel values.
It can be understood that with the processing result in the case of
removing movement blurring using an approximation function in the
actual world made up of a discontinuous function for each region,
change in the pixel value on the edge portion centered on around
x=379, 376 is intensive, movement blurring is removed, so that the
contrast of the image becomes clear, as compared to the processing
result in the case of removing movement blurring by the class
classification adaptation processing.
Also, when movement blurring occurs on an image at the time of a
airplane-shaped object serving as a toy moving in the horizontal
direction as shown in FIG. 411, the comparison between the
processing result in the case of removing the movement blurring
from the image using an approximation function in the actual world
obtained by setting a discontinuous function for each region in the
direction of continuity (image of which the movement blurring
generated with the actual world estimating unit 102 shown in FIG.
369 and the image generating unit 103 shown in FIG. 384 was
removed), and the processing result in the case of removing the
movement blurring from the image using the other method is shown in
A through D in FIG. 412.
That is to say, A in FIG. 412 is the image itself (image prior to
the blurring removal processing) wherein the movement blurring of
the black-frame portion in FIG. 411 occurs, B in FIG. 412 is an
image following the movement blurring being removed from the image
wherein the movement blurring shown in A in FIG. 412 occurred using
an approximation function in the actual world made up of a
discontinuous function set for each region, C in FIG. 412 is an
image taken in a state wherein a subject serving as an input image
is still, and D in FIG. 412 is an image as the processing result of
removing the movement blurring using the other method.
It can be understood that the image (image shown in B in FIG. 412)
of which the movement blurring was removed using the approximation
function in the actual world made up of a discontinuous function
set for each region is a more clear image on the adjacent portion
of "C" and "A" in the drawing, also the regions where characters
exist are displayed more clearly, as compared to the image (image
shown in D in FIG. 412) as the processing result of removing the
movement blurring using the other method. According to this, it can
be understood that fine portions are clearly displayed by the
processing for removing the movement blurring using the
approximation function in the actual world made up of a
discontinuous function set for each region.
Further, when movement blurring occurs on an image at the time of a
airplane-shaped object serving as a toy moving in an oblique
direction (oblique right rising direction) as shown in FIG. 413,
the comparison between the processing result in the case of
removing the movement blurring from the image using an
approximation function in the actual world obtained by setting a
discontinuous function for each region in the direction of
continuity (image of which the movement blurring generated with the
actual world estimating unit 102 shown in FIG. 377 and the image
generating unit 103 shown in FIG. 388 was removed), and the
processing result in the case of removing the movement blurring
from the image using the other method is shown in A through D in
FIG. 414.
That is to say, A in FIG. 414 is the image prior to the blurring
removal processing wherein the movement blurring of the black-frame
portion in FIG. 413 occurs, B in FIG. 414 is an image following the
movement blurring being removed from the image wherein the movement
blurring shown in A in FIG. 414 occurred using an approximation
function in the actual world made up a discontinuous function set
for each region, C in FIG. 414 is an image wherein a subject of the
input image was taken in a still state, D in FIG. 414 is an image
as the processing result of removing the movement blurring using
the other method. Note that the image processed is around a
position appended with a rectangular mark of a thick line in the
drawing of FIG. 413.
As described with reference to FIG. 412, it can be understood that
the image of which the movement blurring was removed using the
approximation function in the actual world made up of a
discontinuous function set for each region is a more clear image on
the adjacent portion of "C" and "A" in the drawing, also the
regions where characters exist are displayed more clearly, as
compared to the image as the processing result of removing the
movement blurring using the other method. According to this, it can
be understood that fine portions are clearly displayed by the
processing for removing the movement blurring using the
approximation function in the actual world made up of a
discontinuous function set for each region.
Further, in the event of removing the movement blurring using the
approximation function in the actual world made up of a
discontinuous function set for each region, upon the upper original
image being input in an oblique direction wherein movement blurring
occurred in the right rising direction shown in A in FIG. 415, the
image such as shown in B in FIG. 415 is output. That is to say, in
the event of the image wherein the pinstriped movement blurring
occurred in the center portion of the original image, the
pinstriped portion becomes a clear image by removing the movement
blurring using the approximation function in the actual world made
up of a discontinuous function set for each region.
That is to say, as shown in A through D of FIG. 412 and A and B of
FIG. 415, the actual world estimating unit shown in FIG. 377 and
the image generating unit 103 shown in FIG. 388 set an
approximation function which estimates the actual world for each
three-dimensional rod-shaped region such as shown in FIG. 391 as a
discontinuous function respectively, and accordingly, it becomes
possible to remove movement blurring which occurs due to movement
not only in the horizontal direction and vertical direction but
also in a oblique direction serving as a combination of those.
According to the above arrangement, the real world light signals
are projected on multiple pixels each having time-space integration
effects, continuity of the image data is detected, of which part of
continuity of the actual world light signals has been lost, the
image data is approximated with a discontinuous function assuming
that the pixel values of the pixels corresponding to a position in
at least one-dimensional direction of the time-space directions of
the image data, corresponding to the continuity of the image data
detected by the image data continuity detecting means, thereby
estimating the function corresponding to the actual world light
signals, and accordingly, it becomes possible to generate high
density pixels used for an enlarged image, and new frame pixels,
and a more clear image can be generated in either case.
Note that the sensor 2 may be a sensor such as a solid-state
imaging device, for example, a BBD (Bucket Brigade Device), CID
(Charge Injection Device), or CPD (Charge Priming Device) or the
like.
Thus, the image processing device according to the present
invention may be provided with input means for inputting image data
made up of multiple pixels acquired by the actual world light
signals being cast upon the multiple detecting elements each having
spatial integration effects via the optical low pass filter, of
which part of continuity of the actual world light signals has been
lost, and actual world estimating means for estimating the light
signals to be cast upon the optical low pass filter considering
that the light signals are dispersed and integrated in at least
one-dimensional direction of the spatial directions by the optical
low pass filter.
The actual world estimating means may be provided, which generates
a function which approximates the real world light signals by
estimating multiple actual world functions assuming that the pixel
value of a pixel of interest corresponding to a position in at
least one-dimensional direction of the spatial directions of image
data is a pixel value acquired by integration in at least
one-dimensional direction of the multiple actual world functions
corresponding to the multiple light signals dispersed in the
spatial direction by the optical low pass filter.
Image data continuity detecting means, which detect continuity of
image data, may be further provided, and based on the continuity
detected by the image data continuity detecting means, the actual
world estimating means may generate a function which approximates
the real world light signals by estimating multiple actual world
functions assuming that the pixel value of a pixel of interest
corresponding to a position in at least one-dimensional direction
of the spatial directions of image data is a pixel value acquired
by integration in at least one-dimensional direction of the
multiple actual world functions corresponding to the optical low
pass filter.
Pixel value generating means may be further provided, which
generates a pixel value corresponding to the pixel having a desired
size by integrating the actual world function estimated by the
actual world estimating means with a desired increment in at least
one-dimensional direction.
Also, computing means for computing image data corresponding to the
light signal when the light signal corresponding to second image
data passes through the optical low pass filter to output the
computed result as first image data, first tap extracting means for
extracting the multiple pixels corresponding to the pixel of
interest within the second image data from the first image data,
and learning means for learning prediction means for predicting the
pixel value of the pixel of interest from the pixel values of the
multiple pixels extracted by the first tap extracting means may be
provided to the learning device, which learns prediction means for
predicting the second image data from the first image data.
Second tap extracting means for extracting the multiple pixels
corresponding to the pixel of interest within the second image data
from the first image data, and features detecting means for
detecting features corresponding to the pixel of interest based on
the pixel values of the multiple pixels extracted by the second tap
extracting means may be further provided to the learning device.
The learning means may be configured so as to learn the prediction
means for predicting the pixel value of the pixel of interest from
the pixel values of the multiple pixels extracted by the first tap
extracting means for each features detected by the features
detecting means.
The computing means may be configured so as to compute the first
image data from the second image data based on the relationship
between phase shift amount, which disperses the light signal of the
optical low pass filter to be processed, and the pixel-to-pixel
distance of the imaging device.
Input means for inputting the first image data acquired by the real
world light signals being cast upon the multiple detecting elements
each having spatial integration effects via the optical low pass
filter, first tap extracting means for extracting the multiple
pixels corresponding to the pixel of interest within the second
image data from the first image data, storing means for storing the
prediction means learned beforehand to predict the second image
data to be acquired by the light signals, which are cast upon the
optical low pass filter from the first image data, and prediction
computing means for predicting the pixel value of the pixel of
interest within the second image data based on the multiple pixels
extracted by the first tap extracting means and the prediction
means may be provided to the image processing device, which
predicts the second image data from the first image data.
Second tap extracting means for extracting the multiple pixels
corresponding to the pixel of interest within the second image data
from the first image data, features detecting means for detecting
the features corresponding to the pixel of interest based on the
pixel values of the multiple pixels extracted by the second tap
extracting means may be further provided to the image processing
device. The prediction means may be learned beforehand so as to
predict the pixel value of the pixel of interest from the pixel
values of the multiple pixels extracted by the first tap extracting
means for each features detected by the features detecting
means.
The prediction means may be learned beforehand so as to predict the
second image data to be acquired by the light signals, which are
cast upon the optical low pass filter from the first image data
computed from the second image data based on the relationship
between phase shift amount, which disperses the light signal of the
optical low pass filter to be processed, and the pixel-to-pixel
distance of the imaging device, being cast directly.
The image processing device according to the present invention may
be further provided with image data continuity detecting means for
detecting continuity of image data made up of multiple pixels
acquired by the real world light signals being cast upon the
multiple detecting elements each having time-space integration
effects, of which part of the continuity of the actual world light
signals has been lost, and actual world estimating means for
estimating the real world light signals by approximating the image
data with a discontinuous function assuming that the pixel values
of the pixels corresponding to a position in at least
one-dimensional direction of the time-space directions of the image
data are pixel values acquired by integration in at least
one-dimensional direction, corresponding to the continuity of the
image data detected by the image data continuity detecting
means.
The actual world estimating means may be configured so as to
generate discontinuous functions divided with a certain increment
in at least one-dimensional direction as a function, which
approximates the real world light signal.
The level of within a certain increment of each discontinuous
function divided with a certain increment may be configured so as
to be a constant value.
The level of within a certain increment of each discontinuous
function divided with a certain increment may be configured so as
to be approximated with a polynomial.
The storage medium storing the program for carrying out the signal
processing according to the present invention is not restricted to
packaged media which is distributed separately from the computer so
as to provide the user with the program, such as a magnetic disk 51
(including flexible disks, optical disk 52 (including CD-ROM
(Compact Disk-Read Only Memory), DVD Digital Versatile Disk),
magneto-optical disk 53 (including MD (Mini-Disk).RTM.),
semiconductor memory 54, and so forth, as shown in FIG. 2, in which
the program has been recorded; but also is configured of ROM 22 in
which the program has been recorded, or a hard disk or the like
included in the storage unit 28, these being provided to the user
in a state of having been built into the computer beforehand.
Note that the program for executing the series of processing
described above may be installed to the computer via cable or
wireless communication media, such as a Local Area Network, the
Internet, digital satellite broadcasting, and so forth, via
interfaces such as routers, modems, and so forth, as necessary.
It should be noted that in the present specification, the steps
describing the program recorded in the recording medium include
processing of being carried out in time-sequence following the
described order, as a matter of course, but this is not restricted
to time-sequence processing, and processing of being executed in
parallel or individually is included as well.
INDUSTRIAL APPLICABILITY
According to the present invention, processing results which are
accurate and highly precise can be obtained, as described
above.
Also, according to the present invention, processing results which
are more accurate and which have higher precision as to events of
the real world can be obtained.
* * * * *